Beruflich Dokumente
Kultur Dokumente
http://avaxho.me/blogs/ChrisRedfield
SIAM's Classics in Applied Mathematics series consists of books that were previousi
allowed to go out of print These books are republished by SIAM as a professional service
because they continue to be important resources for mathematical scientists.
Editor-in-Chief
Robert E O'Malley, Jr., University of Washington
Editorial Board
Richard A. Brualdi, University of Wisconsin-Madison
Herbert B. Keller, California Institute of Technology
Andrzej Z. Manitius, George Mason University
Ingram Olkin, Stanford University
Stanley Richardson, University of Edinburgh
Ferdinand Verhulst, Mathematisch Instituut, University of Utrecht
Bernard Kolman
Drexel University
Philadelphia, Pennsylvania
asm..
Society for Industrial and Applied Mathematics
Philadelphia
Copyright 1972, 1989 by the Society for Industrial and Applied Mathematics.
This SIAM edition is an unabridged, corrected republication of the work first published by
SLAM in 1972.
10 9 8 7 6 5 4 3 2
The work of Bernard Kolman was supported in part by the Air Force Office of Scientific
Research, Air Force Systems Command, United States Air Force, under AFOSR Grant
69-1797.
All rights reserved. Printed in the United States of America. No part of this book may be
reproduced, stored, or transmitted in any manner without the written permission of the
publisher. For information, write the Society for Industrial and Applied Mathematics, 3600
University City Science Center, Philadelphia, PA 19104-2688.
Belinfante, Johan G. F.
A survey of Lie groups and Lie algebras with
applications and computational methods.
The validity of the first paragraph of the Preface to the original edition
Applications of the theory of Lie groups, Lie algebras and their
representations are many and varied. This is a rapidly growing
field through which one can bring to bear many powerful meth-
ods of modern mathematics.
written 18 years ago, has been borne out by the profusion of published
work in this area.
During this period of time we have seen the appearance of several
other presentations of Lie theory directed to various special fields. Another
trend has been the increased emphasis on computational Lie algebra, with
heavy reliance on computers.
We have naturally been gratified and much encouraged by the warm
reception of the first edition. When it became known that a classic edition
was to be published, we received many valuable suggestions for additional
topics to be included. We regret that most of these could not be readily
incorporated without completely changing the outlook, and greatly in-
creasing the size, of the book. We have corrected a few minor misprints
which appeared in the original edition, but the character of the book, espe-
cially its focus on the classical representation theory and its computa-
tional aspects, has not changed.
We thank the many people who wrote us with their comments and
suggestions. We also thank the entire staff of SIAM for their interest and
unfailing cooperation during all phases of this project.
This page intentionally left blank
Acknowledgments
Introduction 3
Chapter 1 Lie Groups and Lie Algebras 5
1.1 The general linear group 5
1.2 Orthogonal and unitary groups 6
1.3 Groups in geometry 7
1.4 The exponential mapping 11
1.5 Lie and associative algebras 12
1.6 Lie groups 13
1.7 Lie algebras of Lie groups 15
1.8 Vector fields 18
1.9 Lie theory of one-parameter groups 19
1.10 Matrix Lie groups 21
1.11 Poisson brackets 24
1.12 Quantum symmetries 26
1.13 Harmonic oscillators 30
1.14 Lie subgroups and analytic homomorphisms 31
1.15 Connected Lie groups 32
1.16 Abelian Lie groups 34
1.17 Low-dimensional Lie groups 35
1.18 The covering group of the rotation group 36
1.19 Tensor product of vector spaces 38
1.20 Direct sums of vector spaces 42
1.21 The lattice of ideals of a Lie algebra 43
1.22 The Levi decomposition of a Lie algebra 44
1.23 Semisimple Lie algebras 45
1.24 The Baker-Campbell-Hausdorff formula 47
Chapter 2 Representation Theory 51
2.1 Lie group representations 51
2.2 Modules over Lie algebras 53
2.3 Direct sum decompositions of Lie modules 56
2.4 Lie module tensor product 57
2.5 Tensor and exterior algebras 59
2.6 The universal enveloping algebra of a Lie algebra 63
2.7 Nilpotent and Cartan subalgebras 65
2.8 Weight submodules 66
2.9 Roots of semisimple Lie algebras 67
2.10 The factorization method and special functions 70
x Contents
2.11 The Cartan matrix 73
2.12 The Weyl group 74
2.13 Dynkin diagrams 76
2.14 Identification of simple Lie algebras 78
2.15 Construction of the Lie algebra A2 79
2.16 Complexification and real forms 80
2.17 Real forms of the Ue algebra Al 83
2.18 Angular momentum theory 86
Chapter 3 Constructive Methods 91
3.1 Raising and lowering subalgebras 91
3.2 Dynkin indices 93
3.3 Irreducible representations of Al 95
3.4 The Casimir subalgebra 97
3.5 Irreducible representations of A2 99
3.6 Characters 101
3.7 Computation of the Killing form 103
3.8 Dynkin's algorithm for the weight system 106
3.9 Freudenthal's algorithm 109
3.10 The Weyl character formula 111
3.11 The Weyl dimension formula 114
3.12 Characters of modules over the algebra A2 116
3.13 The Kostant and Racah character formulas 117
3.14 The Steinberg and Racah formulas for Clebsch-Gordan
series 119
3.15 Tensor analysis 122
3.16 Young tableaux 124
3.17 Contractions 128
3.18 Spinor analysis and Clifford algebras 131
3.19 Tensor operators 137
3.20 Charge algebras 141
3.21 Clebsch-Gordan coefficients 145
Bibliography 149
Index 159
A Survey of
Lie Groups and Lie Algebras
with Applications and
Computational Methods
This page intentionally left blank
INTRODUCTION
Since many of the central ideas of Lie theory arose in the study of
geometry and linear algebra, it is fitting to begin with a review of some topics
in vector space theory so that we can begin to talk about some of the essential
tools, such as the Lie algebra and tensor algebra constructions. It is also
instructive to examine the classical linear Lie groups before getting involved
with the rather technical general definition of a Lie group.
We recall that a linear mapping a: V^ - V2 from a vector space Vl into a
vector space V2 is a mapping which preserves vector addition and scalar
multiplication:
and
The general and special linear groups may then be regarded as groups of
matrices, the group operation being matrix multiplication. These groups of
5
6 Lie Groups and Lie Algebras 1.2
matrices can be defined for any field of scalars such as, for instance, the real
number field R or the complex number field C. Thus, in particular we obtain
the group of all nonsingular complex n x n matrices, called the complex
general linear group GL(n, C).
The real general linear group GL(n, R), consisting of all real nonsingular
n x n matrices, may be regarded as a subgroup of the complex general linear
group. The complex special linear group SL(n, C) is the subgroup of GL(n, C)
consisting of matrices with determinant one. The intersection of these two
subgroups is the real special linear group,
The special linear groups are sometimes also called unimodular groups. The
reader should be warned however that in abstract harmonic analysis the
same term, unimodular group, is used in a completely different sense [152].
The real orthogonal group may be considered as the group of all rigid motions
which leave a given point fixed. The rotation group is the special or proper
real orthogonal group,
we can compute
Since the trace of an operator is also equal to the sum of its eigenvalues,
counting multiple roots of the characteristic equation, it does not matter
which basis we use here. The set linF (K) of all linear operators on an n-di-
mensional vector space Fover a field F is itself an 2-dimensional vector space
over F. Then TrK(a/?) defines a bilinear form on the vector space lin F (K).
This type of symmetric bilinear form is important in the study of the structure
of Lie groups and Lie algebras.
The introduction of a symmetric or antisymmetric bilinear form in a
vector space leads naturally to the idea of orthogonality of vectors. Vectors
x and y in V are said to be orthogonal if (x, y) = 0. Much of the theory of
symmetric bilinear forms also applies to antisymmetric bilinear forms, and
we may discuss the two cases simultaneously.
For a symmetric or antisymmetric bilinear form, the orthogonality of
vectors is a symmetric relation ; if x is orthogonal to y, then y is also ortho-
gonal to x. A nonzero vector x is said to be isotropic, or a null vector, if it is
orthogonal to itself, (x, x) = 0, and anisotropic if (x, x) ^ 0. Isotropic vectors
are of course not present in Euclidean geometry, but they often occur in more
general situations. In special relativity theory, isotropic vectors are called
light-like because light rays travel in straight lines along the directions of such
vectors. For symmetric bilinear forms over the complex number field,
1 .3 Groups in Geometry 9
isotropic vectors occur in any vector space having dimension ^2. If we are
dealing with antisymmetric bilinear forms over the real or complex number
field, then every vector is isotropic.
For the discussion of the orthogonal and symplectic groups, we shall
restrict our attention to nonsingular bilinear forms. A bilinear form is
nonsingular if the only vector orthogonal to every other vector is zero, that is,
if (x, y) = 0 for all y in V implies that x = 0. The bilinear form (x, y) is non-
singular if and only if its matrix ft is invertible, which means that det ft ^ 0.
It should be noted that whether or not a bilinear form is nonsingular, its
restriction to a subspace need not be nonsingular. Thus the notion of non-
singularity also leads to a classification of the subspaces of a vector space, and
this is often useful in the structure theory of vector spaces equipped with
bilinear forms.
An orthogonal linear operator a with respect to a nonsingular symmetric
bilinear form is a linear operator for which
(ax, ay) = (x,y)
holds for all x and y in K If x belongs to the kernel (null space) of an orthog-
onal linear operator a, then we have ax = 0, and hence for all y we find
(x, y) = (ax, ay) = 0. Since the bilinear form is nonsingular, x = 0 and a is
one-to-one, and if V is finite-dimensional, this further implies that a is
invertible. Thus the set of orthogonal linear operators in this case forms a
group, which we call the orthogonal group with respect to the given symmetric
bilinear form.
We obtain the matrix groups mentioned earlier by introducing bases for
vector spaces equipped with nonsingular bilinear forms. The matrix ft
describing the bilinear form depends on the choice of the basis ev , , en.
If ft is the matrix of this same bilinear form with respect to another basis
e\, , e'n, where e\ = Xj^Vj' then ft' is related to ft by the congruence
transformation
the condition that its inverse is its transpose. Therefore, we obtain just one
orthogonal group for each dimension in the complex case, namely the group
0(ii, C).
In the real case, if det ft ^ 0, we can find a matrix a so that /?' is diagonal
and its diagonal elements are 1. The signature (p, q), where p and q are the
numbers of + 1's and 1's respectively on the diagonal, is an invariant of the
real symmetric bilinear form. In the special case where p = n and q = 0, the
real symmetric bilinear form is said to be positive definite because (x, x) > 0
for all x ^ 0, and we obtain the ordinary real orthogonal group O(n, R). In
the general case, the group O(p, q ; R) which we obtain may be described as an
orthogonal group with an indefinite metric of signature (p, q) in a space of
dimension n = p + q. An example of this is the Lorentz group 0(3, 1 ; R),
which is just the orthogonal group of the Minkowski space-time R 4
(+ + + -)
The definition of the symplectic groups is somewhat similar to that of the
orthogonal groups. Just as the orthogonal groups consist of linear operators
which leave invariant a symmetric nonsingular bilinear form, the symplectic
groups consist of linear operators which leave invariant an antisymmetric
nonsingular bilinear form. A symplectic linear operator a is an operator
satisfying
Taking intersections with SL(n, C) gives nothing new since every symplectic
linear operator already has determinant one [10]. The symplectic groups have
been applied to various problems in nuclear physics [121]. There are many
other matrix groups, and more extensive catalogues of them can be found
in the literature [120].
We may also use the exponential map to get linear conditions in the
case of the orthogonal, unitary and symplectic groups. The transpose of the
exponential of a matrix is the exponential of its transpose, and the inverse of
the exponential of a matrix is the exponential of its negative. Hence, a
sufficient condition for exp (A) to be orthogonal is that A be antisymmetric.
Similarly, the complex conjugate of exp (A) is the exponential of the complex
conjugate of A. Hence, exp (A) is unitary when A is equal to the negative of its
transposed complex conjugate. Such a matrix is called anti-Hermitian.
Finally, we note that exp (^4) is symplectic if A satisfies ATJ + JA = 0, which
is again a linear condition.
In summary, the exponential mapping allows us to replace the classical
matrix Lie groups by sets of matrices which satisfy various linear conditions.
Such sets of matrices are of course closed under taking arbitrary real linear
combinations, and we may view these sets as vector spaces. These sets of
matrices are not closed under ordinary matrix multiplication. However, if A
and B are matrices which are, respectively, antisymmetric, anti-Hermitian or
traceless, then AB BA also has that property. Hence, the sets of matrices
with which one has to deal are not only vector spaces, but are also closed under
an additional operation, and one says that they form a Lie algebra. The study
of these Lie algebras, as well as others just like them, forms the central core
of the Lie theory.
Before studying the Lie algebras of the classical matrix groups in detail,
it is useful to discuss Lie algebras in general from a slightly more formal view-
point. In addition, we wish to discuss some relations existing between Lie
algebras and associative algebras. In general, any vector space endowed with
a bilinear vector multiplication is called a nonassociative algebra [207]. That
is, a nonassociative algebra consists of a vector space V together with any
bilinear mapping V x V -> V. The product of two vectors x, y in V then is
another vector xy in V. If the vector multiplication is associative, so that
(xy)z = x(yz), then we speak of an associative algebra. For example, the set
Imp (V) of all linear operators on a vector space Kover a field F is an associa-
tive algebra under the usual multiplication of linear operators. This same
vector space lin F (F) also has another algebra structure defined on it by
taking the product of linear operators a and /? to be the commutator ctfi /fe.
To distinguish this product from the usual product, we use a special square
bracket notation [a,)3] for it. The resulting algebra is a nonassociative
algebra of a certain special type called a Lie algebra.
Abstractly, a Lie algebra L is a vector space equipped with a product
[x, y] satisfying certain axioms [73], [82], [83], [135], [228]. We shall continue
to use the bracket notation for products when we deal with any Lie algebra.
One of the axioms for a Lie algebra is that the product [x, y] be bilinear, that
1.6 Lie Groups 13
is, linear in x and y separately. We also assume that the Lie product is-anti-
commutative,
holds for all vectors x, y, z in the Lie algebra. The Jacobi identity plays for
Lie algebras the same role that the associative law plays for associative
algebras. While we can define Lie algebras over any field, we shall in practice
only consider Lie algebras over the real and complex fields.
Undoubtedly the most familiar abstract Lie algebra is the real three-
dimensional vector space with the vector cross product as multiplication.
Obviously the vector cross product Lie algebra can also be defined for a
three-dimensional vector space over any field. Anticommutativity of the
vector cross product is familiar, and the Jacobi identity for cross products
can be verified by a direct computation, using
For Imp (V) with the commutator multiplication it is likewise easy to verify
directly that the Lie algebra axioms are satisfied. The procedure used to give
linF (V) a Lie algebra structure can be extended to any associative algebra. A
Lie product can be defined in any associative algebra by the commutator
[x, y] = xy yx, making it a Lie algebra. Here the anticommutativity axiom
is clearly satisfied, and the reader can check that in this case the Jacobi identity
for the commutator follows from the associative law for the ordinary product.
There is another and perhaps more revealing way in which Lie algebras
arise in the study of algebras. A derivation dot a nonassociative algebra A is a
linear operator on A satisfying the formal analogue of the Leibniz rule for
differentiating a product,
for all elements x and y in A. If p(x) is any polynomial in x, then the differential
operator p(x) d/dx is an example of a derivation of the algebra of polynomials
in x. In fact, the concept of a derivation is just an abstraction of the idea of a
first order differential operator. The set of all derivations on a nonassociative
algebra A is clearly a subspace of the linear operator algebra linF (A).
Although the product of derivations is in general not a derivation, the
commutator d^d2 - d2dl of two derivations is again a derivation. Thus the
set of all derivations of a nonassociative algebra is a Lie algebra, which we
call the derivation algebra of the given nonassociative algebra.
The general theory of Lie groups seeks to unify and to extend the dis-
cussion of the various classical groups. The extension of the theory comes
about by treating not only matrix groups, but other types of groups as well.
14 Lie Groups and Lie Algebras 1.6
Basically, a Lie group is the structure which naturally results when analytic
machinery is coupled with abstract group theory. Curvilinear coordinates,
derivatives, and power series become available as tools for the study of the
resulting structure. Technically, a Lie group is first of all required to be a
topological group, that is, it must be possible to assign a topology under which
the group operation and taking inverses are both continuous functions
[129], [164], [192]. The general concept of a topological group already leads
to a very rich theory in which some of the methods of calculus, especially
integration theory, can be applied. To make use of the full power of all the
methods of calculus, including differentiation, we must introduce an analytic
structure as well as a topology. A proper discussion of the analytic structure
of a Lie group requires the concept of an analytic manifold [38], [42], [54],
[64].
Historically, the concept of an analytic manifold grew out of the theory
of Riemann surfaces in complex variable theory [238]. The theory of differen-
tiate and analytic manifolds has since become the basis for modern
differential geometry [127], [230], The idea of defining a Lie group in terms
of analytic manifolds did not gain widespread acceptance until almost half
a century after Lie's original work [163]. Moreover, it turned out that any
connected real analytic manifold may always be analytically embedded in a
Euclidean space, so that it appears after all that manifold theory is somewhat
of a luxury [171]. Therefore, instead of using the intrinsic definition, we shall
regard a connected analytic manifold more intuitively as a smooth surface in a
Euclidean space of suitable dimensions. In general, an analytic manifold will
consist of several pieces if it is not connected. About each point of the
manifold there is required to be an open set with all points within this open
set being located by a curvilinear coordinate system. It is also required that
the assignment of coordinates to points be a continuous mapping with a
continuous inverse. Such an open subset of a manifold is called a coordinate
neighborhood or coordinate patch, and the coordinate system is said to give a
chart of this neighborhood. To avoid singular points like the poles on a
Mercator projection map, we use many charts, forming an atlas which covers
the manifold, each point of the manifold belonging to at least one chart. In
general, these charts may overlap somewhat so that two coordinate neighbor-
hoods may have a point in common. Overlapping charts are required to be
analytically related, meaning that for any common point, the coordinates in
each system are analytic functions of those in the other. A function defined
in Euclidean space R" is analytic at a point if it can be expressed as a con-
vergent Taylor series in some neighborhood of the point.
A Lie group is a group which is also an analytic manifold and for which
the group operations are analytic functions. Because of the analytic structure,
each element of the group is specified by some curvilinear coordinate system.
The multiplication of elements of a Lie group is required to be analytic in
terms of these coordinates. Thus if x and y are points of the group, and if
z = x o yt where o denotes the group operation, then we require that the
1.7 Lie Algebras of Lie Groups 15
The whole atlas of charts of a Lie group can be generated from the
charts at the identity element. If U is a coordinate neighborhood of an
element g, then the set g-1U obtained by letting g" 1 act on all elements of
U will contain a coordinate neighborhood of the identity element. By this
procedure, every coordinate neighborhood can be transported back to the
identity, and it suffices to study what happens close to the identity. For
convenience, coordinates in a neighborhood of the identity will be chosen
so that the coordinates of the identity are zero. If x is a member of the group,
we denote its coordinates by x'. If x <> y = 2, we have z'(x! , , x", y 1 , - - , / ) ,
by hypothesis, an analytic function which can be expanded in a convergent
Taylor series about the origin. Thus,
Since cljk = clkj, the quadratic term drops out, proving our assertion.
1.7 Lie Algebras of Lie Groups 17
Thus the tensor c'jt determines the predominant part of the commutator.
The local structure of a Lie group can be summarized conveniently in
terms of the tangent space to the Lie group at the identity element. Let
xl , , x", as before, be the coordinates of an element x in a neighborhood
of the identity in the Lie group. Let v(x1, , x") be the position vector of this
point with respect to the origin of the surrounding Euclidean space. Then the
vectors
form a basis for the tangent space at the identity element of the Lie group.
The coordinates x', y\ zl of the points of the Lie group close to the origin,
used in the previous expressions, can be regarded as components of vectors
in the tangent space at the identity. That is, we construct the vectors
The properties of this vector multiplication follow from the properties of the
structure constants c'jk. It follows from the definition
definition of the structure constants
that they are antisymmetric in their lower indices,
Therefore, we have
18 Lie Groups and Lie Algebras 1.8
We might ask what further properties c'jk must have. Substituting the Taylor
series for Lie group multiplication into the equation
and equating the corresponding terms, one obtains expressions relating the
c'jk to coefficients of higher order terms. By permuting the indices cyclically
and summing, one can eliminate the coefficients of the higher order terms,
obtaining the relation
Therefore, our vector multiplication also satisfies the Jacobi identity, and we
have a Lie algebra. The tangent vector space at the origin, with this multiplica-
tion, is called the Lie algebra of the Lie group. In the older literature the Lie
algebra was called the infinitesimal group. The components of the tensor c'jk
are called the structure constants of the Lie algebra with respect to the basis
e
i j ' ' ' > e -
The local structure of a Lie group, the structure in a sufficiently small
neighborhood, is completely determined by the Lie algebra. This is of great
importance in applications, for when properties of a local nature are being
studied one need only consider the Lie algebra.
a vector field. Thus the set of all vector fields may be regarded as a vector
space. If we set
then
with the differential equation, then the equation is invariant under the group
if and only if
which frequently arises. In many applications one deals with equations which
are invariant with respect to changes of the units used for measuring the
quantities which appear in them. Then the process of finding u and v can be
carried out simply by performing a dimensional analysis of the units for these
quantities [34]. When the differential equation is formulated in terms of u
and v its order is one less than the order of the original equation. If the
dimension of the Lie group under which the differential equation remains
invariant is greater than one, the integration procedure can again be simpli-
fied by introducing canonical coordinates. These canonical coordinates and
the integration procedure depend in general on the structure of the Lie
algebra of the group of transformations and on the relative dimensions of the
group and the orbits. The dimension of an orbit may well be less than the
dimension of the group itself. For instance, the group 50(3, R) is a three-
dimensional group of transformations on three-dimensional Euclidean space,
while its nontrivial orbits are the spheres which are two-dimensional.
In addition to giving methods of reducing differential equations in-
variant under a Lie group to quadratures, Lie also showed how to find the
differential equations invariant under a given group. The problem of finding
invariant differential equations has many applications. For example, in
relativistic quantum field theory, a basic problem is that of finding all wave
equations invariant under the Lorentz group [66]. The problem simplifies
somewhat for the case of zero-mass wave equations because in that case
relativistic invariance implies in variance under the bigger group of conformal
transformations [95].
The problem of finding differential equations invariant under a given
Lie group also comes up in quite different fields. We mention here some work
by Hoffman on pattern recognition in visual perception [125]. On the basis
of physiological studies of electrical activity in the brain, he postulates that the
visual integrative process is based on a first order differential equation, with
electrical patterns in the brain embodying information about the isoclines of
the differential equation. In order for recognition of a pattern as an entity
to occur, as required by the Gestalt theory, this differential equation must be
invariant under the group of transformations which leave the pattern
recognizable. Hoffman determined the class of differential equations invariant
under a Lie group which includes translations, rotations and magnifications.
The results of this analysis are used to explain and interpret various visual
phenomena, including developmental dyslexia, the whirling spirals evoked
under flicker, the alpha rhythm and its desynchronization, and Mackay's
complementary after-images.
Thus we have
1.10 Matrix Lie Groups 23
Therefore, the structure constants for the general linear group are given by
The tangent space of the general linear group can be identified with the space
of all n x n matrices, including also those which are not invertible. If x and y
are tangent vectors (matrices) with components xu and ym", then we have
[x, y] = xy yx, where the usual matrix multiplication is meant on the
right side. We have thus determined the structure of the Lie algebra of the
general linear group of any order. The real Lie algebra of GL(n, R) is denoted
by gl(n, R), and the real Lie algebra of GL(n, C) is denoted by gl(n, C). In
general, the real Lie algebra of a Lie group may be denoted by using the
lower case letters corresponding to the letters used to denote the Lie group.
The Lie algebras of many of the classical matrix groups can be determined
by means of the exponential mapping. If each matrix in some neighborhood
of the identity in a matrix group can be expressed as exp (A) for A in some
linear subspace of the space of square matrices, then this subspace constitutes
the tangent space to the group. When endowed with the commutator
multiplication, [A, B] = AB BA, this linear subspace of square matrices
is the Lie algebra. For example, the real Lie algebra s/(n, C) of the special
linear group consists of the matrices with trace zero. The real Lie algebra
o(n, C) of the complex orthogonal group consists of all complex antisym-
metric matrices of order n, while the Lie algebra o(n, R) consists of the real
ones. Similarly, the real Lie algebra u(ri) of the unitary group consists of the
anti-Hermitian matrices. The tangent space to the intersection of two
manifolds at a point is the intersection of their tangent spaces. Hence the Lie
algebras of the matrix groups obtained by intersecting GL(n, C), GL(n,R),
SL(n, C), 0(n, C) and U(n) are simply the intersections of their Lie algebras.
For our final example, let us compute the Lie algebra so(3, R) of the
rotation group in ordinary three-dimensional Euclidean space. Our first
problem, of course, is to parametrize the matrices of the rotation group. We
can give a general formula for the matrix [atj] corresponding to an arbitrary
rotation through an angle
element of the group S0(3, R). The tangent space at the identity has as a basis
the three matrices ek having (y)-entry
Thus,
and
If we identify the vectors e-t with the usual unit vectors of vector analysis, we
recognize this to be the ordinary vector cross product. Therefore, the Lie
algebra so(3, R) of the rotation group is isomorphic to the three-dimensional
real Euclidean space with the vector cross product as the Lie multiplication.
Of course, we could equally well have arrived at the cross product algebra in
this case by using the exponential mapping.
Lagrange's equations have the same form in any coordinate system because
they are the equations for the extremal curves which minimize a certain
integral.
To obtain an even more symmetric formulation for classical mechanics,
we must pass from this Lagrangian description to the Hamiltonian descrip-
tion via the Legendre transformation. We introduce the generalized momenta
PJ associated with the corresponding coordinates qj by
are not valid for an arbitrary system of coordinates in phase space. They
hold only for certain sets of canonical coordinates which split into two sets, a
set of position coordinates and the corresponding momentum coordinates.
By a dynamical variable, we mean any infinitely differentiable function
of the coordinates, momenta and time. The dynamical variables form an
infinite-dimensional real Lie algebra with the vector multiplication
It follows that if A and B are conserved, then the Poisson bracket of A and B
is conserved. Also any real linear combination of conserved quantities is
conserved, and hence the set of all conserved dynamical variables forms a
subalgebra of the Lie algebra of all dynamical variables.
for all vectors HP and <i> in the Hilbert space. We also have to deal with anti-
unitary operators, which are not linear operators at all, but are instead
antilinear. An antilinear operator A in Hilbert space is a mapping which
preserves vector addition, but satisfies
for all complex numbers c and vectors 4/. Here c* denotes the complex
conjugate of the number c. An antiunitary operator A is an invertible anti-
linear operator which satisfies
If, conversely, every unit vector also corresponds to some physical state,
we say that there are no superselection rules [227]. The relation between pure
states and unit vectors is not one-to-one since it is assumed that unit vectors
differing only by a phase factor will correspond to the same state. In the
absence of superselection rules, we can describe the relation between Hilbert
space and pure states as a mapping of the unit sphere in Hilbert space onto
the set of pure states. If O and T are unit vectors, then |(<I>, )|2 is interpreted
as the probability that a system prepared in the pure state corresponding to
<X> will be found upon measurement to be in the pure state corresponding to
*F. Note that states which are indistinguishable by these transition probabili-
ties are considered to be identical states.
A physical symmetry is an onto mapping of pure states to pure states
which preserves all transition probabilities. The assumption about in-
distinguishability implies that physical symmetries must also be one-to-one
as mappings and hence have inverses. The products and inverses of physical
symmetries are again symmetries, and hence the physical symmetries form a
group. Wigner discovered that physical symmetries correspond to unitary or
antiunitary operators in Hilbert space [19], [249]. This correspondence is
again unique only to within phase factors. The group generated by these
unitary and antiunitary operators will therefore in general be an extension
of the physical symmetry group. Technically, the physical symmetry group
will be isomorphic to the quotient group of this group of operators modulo
the group of phase factors [211].
Although the descriptions of classical and quantum symmetries appear
to be very different, the classical description must of course be a limiting case
of the quantum description. There actually is a close relation between the
classical and quantum descriptions of dynamics, which historically played an
important role in developing quantum mechanics via the correspondence
principle. The correspondence principle was used to guess the quantum
analogues of classical dynamics in the early days of quantum mechanics.
Any such method of assigning quantum analogues to classical dynamical
variables may be called a quantization procedure. Starting with the methods
proposed in 1926-1927 by P. A. M. Dirac, J. von Neumann and H. Weyl, a
large variety of quantization procedures have been studied over the years
[210]. One of the most useful of these quantization methods, due to Dirac,
makes use of Lie algebraic ideas. The Dirac correspondence principle relates
the Poisson brackets of classical mechanics to quantum mechanical com-
mutators [75], [76], [79]. In this theory the classical Lie algebra of dynamical
variables is related to a Lie algebra of operators in Hilbert space in quantum
mechanics. By using this correspondence principle, we can often discuss
symmetries of analogous systems in a parallel fashion. For example, the
Kepler problem of planetary motion and the hydrogen atom are analogous
systems and so have the same symmetries (the four-dimensional orthogonal
group), which we discuss further later on.
1.12 Quantum Symmetries 29
and
and
with the Hamiltonian. These latter commutation relations imply that if ij/
is an eigenvector of H with eigenvalue E, then ax\}/ is either zero or an eigen-
vector with eigenvalue E 1, while a* if/ is an eigenvector with eigenvalue
E + 1. Thus the harmonic oscillator has an infinite series of equally spaced
levels.
Both the Hamiltonian and the commutation relations are invariant
under a unitary transformation,
appears at first sight to be symmetric under the orthogonal group 0(2n, R).
However, one must remember that Hamilton's equations hold only for
certain canonical systems of coordinates in phase space. In the case of linear
transformations of the coordinates (ql, , q", pj , , pj in phase space,
this requirement limits us to symplectic linear transformations [65] . In view
of the isomorphism between U(n) and Sp(n, R) fl 0(2n, R), we see that the
symmetry group of a system of n identical classical harmonic oscillators is
again the unitary group [120].
group. The Lie algebra of a Lie subgroup of a Lie group may be naturally
identified with a subalgebra of the Lie algebra of the original group. Con-
versely, every subalgebra of the Lie algebra of a Lie group may be identified
with the Lie algebra of a connected Lie subgroup. Every closed subgroup of a
Lie group is a Lie subgroup. However, a Lie subgroup of a Lie group need
not be a closed subgroup. From the theory of topological groups it is known
that every open subgroup is closed, and hence open subgroups of Lie groups
are also Lie subgroups [64].
Similarly, a homomorphism between Lie groups in general need not be
an analytic mapping. A homomorphism between Lie groups is called an
analytic homomorphism if the coordinates of the image of a point are analytic
functions of the coordinates of the point. Analytic homomorphisms of Lie
groups induce homomorphisms of the corresponding Lie algebras. The
homomorphism induced by a composition of analytic homomorphisms
is the composition of their induced homomorphisms. This relation between
homomorphisms of Lie groups and Lie algebras can be used to translate
many results about Lie groups into related results about their corresponding
Lie algebras. For example, the inclusion mapping for a Lie subgroup is an
analytic homomorphism, and hence induces the corresponding embedding
for their Lie algebras noted above. It is a remarkable fact that any con-
tinuous homomorphism of Lie groups is also analytic [124].
If H is a closed normal subgroup of a Lie group G, then we may regard
the quotient group G/H as a Lie group in a natural way, and the natural
projection is analytic since it is known to be continuous. Finally, as one might
expect, a form of the fundamental theorem of homomorphisms holds for Lie
groups, and it is related to the corresponding theorem for their Lie algebras.
The kernel of an analytic homomorphism of a Lie group is a closed normal
subgroup. The kernel of the induced Lie algebra homomorphism is the Lie
algebra of the kernel of the group homomorphism.
The study of a Lie group G does not quite reduce to studying the con-
nected identity component G0 and the discrete quotient group G/G0. One
also needs to know the action of certain inner automorphisms restricted to
the identity component to obtain the complete structure of a Lie group.
Frequently, for Lie groups of practical importance, the structure of the
quotient group is extremely simple, and knowledge of it together with
knowledge of the structure of the identity component is all that is needed to
completely determine the group structure.
For the second stage of the reduction of the global structure of a Lie
group, our analysis is directed toward the structure of the identity component
G0, which is a connected Lie group. The Lie algebra of the original Lie group
is isomorphic to the Lie algebra of the identity component. This Lie algebra
is also isomorphic to the Lie algebra of a simply-connected Lie group, which
is called the universal covering group of the connected Lie group G 0 . The Lie
group G0 is the homomorphic image of the simply-connected universal
covering group determined by the Lie algebra. The kernel of this homo-
morphism is a discrete normal subgroup of the universal covering group
called the fundamental group or Poincare group [36], [200], [205]. Each point
of this discrete fundamental group corresponds to a set of loops in the
connected group G 0 . Two such loops correspond to the same point of the
kernel if and only if one can be continuously transformed into the other.
We may illustrate some of the results of the global structure theory for
Abelian Lie groups. A Lie group is Abelian if
for all elements a and b of the group. The simplest Abelian Lie groups are the
vector groups. A vector group is a real vector space R", regarded as an
Abelian Lie group under the operation of vector addition, using any
rectilinear coordinate system to provide a chart. Vector groups are simply-
connected, and conversely, every simply-connected Abelian Lie group is
isomorphic to a vector group. An Abelian Lie algebra is a Lie algebra whose
structure constants are all equal to zero, so that
for all elements x and y of the Lie algebra. Note that any real vector space may
be made into an Abelian Lie algebra by introducing this trivial multiplication.
We can now use the theory of the universal covering group to classify
all the connected Abelian Lie groups. The Lie algebra of any Abelian Lie
group is an Abelian Lie algebra, and we may regard the Lie algebra itself as
a vector group. Obviously, a vector group is its own tangent space, and so
with the trivial multiplication, is its own Lie algebra. Hence, every connected
1.17 Low-Dimensional Lie Groups 35
Abelian Lie group has a vector group as its universal covering group. Every
connected Abelian Lie group is then a homomorphic image of a vector group,
the kernel being a discrete subgroup. The discrete subgroups of a vector
group are crystallographic lattices of vectors of the form
where the n, run over all integers and the e, are fixed linearly independent
vectors.
Thus every connected Abelian Lie group is a direct product of lines
and circles. The one-dimensional simply-connected Abelian Lie groups are
isomorphic to the additive group of the real line. Any discrete subgroup of the
real line consists of equally spaced points, all the integer multiples of a fixed
real number. Identifying points of the line modulo this discrete subgroup is
equivalent to winding the line about a circle of circumference equal to the
spacing between the points of the subgroup. If k is the least positive member
of the discrete subgroup, then
maps the real line homomorphically onto the circle group, and has the
discrete subgroup as kernel. Thus, the circle group has the real line group as
its universal covering group, and the discrete fundamental group of the circle
group is isomorphic to the additive group of integers. In short, the line and the
circle are the only connected one-dimensional Abelian Lie groups. The only
two-dimensional connected Abelian Lie groups are the plane, the cylinder
and the torus. Note that the plane group may also be regarded as the additive
group of all complex numbers, and the cylinder group as the multiplicative
group of the nonzero complex numbers. For higher dimensions, the results
are similar, there being exactly n + 1 different connected w-dimensional
Abelian Lie groups.
Many Lie groups may be realized as matrix Lie groups. We recall that
analytic subgroups of a Lie group correspond to subalgebras of its Lie
algebra, and the Lie algebra may be taken to be the tangent space at the
identity. The tangent space of a matrix Lie group is some subspace of the
tangent space of the general linear group, the latter being identified with the
space of all square matrices of a given order. The Lie product in the tangent
space of a matrix Lie group is always the commutator, just as it is for the
general linear group.
To illustrate this, let us consider the real affine group consisting of all real
matrices of the form
36 Lie Groups and Lie Algebras 1.18
We may think of this real affine group as the identity component of the group
of inhomogeneous linear transformations on the real line [11], [12]. Since the
underlying manifold of the real affine group is a plane with coordinates (a, b),
the group is a simply-connected Lie group. A basis for the tangent space
consists of the matrices
In fact, any non-Abelian two-dimensional real Lie algebra has this structure
for an appropriate choice of basis. Since this affine Lie algebra is the only
non-Abelian two-dimensional one, any connected two-dimensional non-
Abelian Lie group is isomorphic to the real affine group modulo a discrete
normal subgroup. By writing out the form of a conjugate class, it is easily
seen that no proper normal subgroup can be discrete. The real affine group
is therefore the only connected non-Abelian two-dimensional Lie group.
Thus, the only possible connected two-dimensional Lie groups are the plane,
the cylinder and the torus. The real Cartesian plane R2 admits two group
structures making it a Lie group, one Abelian and one non-Abelian. The
torus and the cylinder admit only an Abelian Lie group structure.
For higher dimensions, finding all connected Lie groups becomes a
much more complicated problem. A part of the solution, finding all the
possible real Lie algebras of a given dimension, has been studied [174] , [175] ,
[176], [177]. We shall say more about this part of the problem later on.
ff
where a, b, c and d are real numbers, and 1 is the identiy matrix. Since the
determinant of such a matrix is a2 + b2 + c2 + d2, we see that the group
1.18 The Covering Group of the Rotation Group 37
manifold of SU(2) can be identified with the unit sphere in a real four-
dimensional Euclidean space with coordinates (a,b,c,d). Thus the special
unitary group SU(2) is a simply-connected three-dimensional Lie group [54].
For any Q in 5 7(2), we define a 3 x 3 matrix [atj] by
This identity can be used to verify that the mapping of SU(2) onto 50(3, R) is
a group homomorphism. By the fundamental homomorphism theorem,
the rotation group 50(3, R) is isomorphic to the quotient group of SU(2)
modulo its center, the discrete subgroup consisting of the identity and its
negative. Incidentally, since the group 5(7(2) is homeomorphic with the
unit sphere in 4-space, it is clearly compact, and the rotation group 50(3, R),
being a continuous image of 5L^(2), is also compact.
The relationship between the groups 50(3, R) and 517(2) can also be
made explicit in the following alternative manner. The elements of 50(3, R)
correspond in a one-to-one fashion with the rotations of a sphere about its
center. The points of the sphere other than the "north pole" may be identified
with the complex plane by the usual projection from the "north pole" onto a
plane tangent to the sphere at the "south pole." Under this projection, each
rotation of the sphere induces a corresponding mapping of the complex
plane onto itself, a complex function of a complex variable. The complex
functions found in this way are certain linear fractional transformations of
the form
where a, /?, y and d are complex numbers and a<5 /?y ^ 0. The composition
of two such linear fractional transformations is again a linear fractional
transformation. Moreover, the condition a<5 /?y ^ 0 ensures that the
transformation has an inverse, which is again a linear fractional transforma-
tion.
The linear fractional transformations form a group, and the rotation
group is isomorphic to a subgroup of this group. Obviously /(z) is unchanged
if all the coefficients a, /?, y, 6 are multiplied by a common factor, so by
introducing an appropriate factor we can always make ad yf$ 1. Since
only the square of a common factor enters into the expression a<5 y/?, its
38 Lie Groups and Lie Algebras 1.19
and its negative, both with determinant one. The linear fractional transforma-
tion corresponding to a product of two such matrices is the composition of
the linear fractional transformations corresponding to the matrices being
multiplied. Thus we have a homomorphism of the group SL(2, C) onto the
group of linear fractional transformations, and the kernel of this homo-
morphism consists of the identity matrix and its negative. Not every linear
fractional transformation, however, corresponds to a rotation of a sphere.
In fact, the group of linear fractional transformations is isomorphic to the
identity component of the Lorentz group. A linear fractional transformation
corresponds to a rotation of the sphere if and only if it corresponds to a
unitary matrix. That is, its matrix must have the form
where a, b, c and d are real numbers. The condition a<5 /?y = 1 is precisely
the condition that the vector (a, b, c, d) lie on the unit sphere in 4-space.
Thus again we obtain the homomorphism of SU(2) onto S0(3, R) as before.
Diagram 1 summarizes the interrelations between the various groups
discussed here.
The study of the local structure of Lie groups, based on the theory of
Lie algebras, makes heavy use of linear algebra and vector space concepts.
For the complexification process, as well as in connection with the theory
of bilinear forms and elsewhere, it is convenient to use a basic concept of
vector space theory, the tensor product. We may recall that in the
discussion of the geometric concept of orthogonality and in the definition of
1.19 Tensor Product of Vector Spaces 39
associative and Lie algebras, we already had to deal several times with bilinear
functions.
There are actually several different, but equivalent, ways to define the
tensor product of two vector spaces V1 and V2 . The best way is to give a
universal definition, defining the tensor product by means of its properties.
Another way, which assures the existence of the object being defined, is to
give a completely constructive definition, and this is what we shall do first.
We temporarily use the nonstandard but rather suggestive notation v1 x v2
for the ordered pair (i^ , v2) having for its first element a vector vl in V^ and
for its second element a vector v2 in V2 . The set of all such ordered pairs
can be regarded as the basis of an infinite-dimensional vector space consisting
of all their finite formal linear combinations. A typical vector in this space
is a vector
and
If {vlt} is a basis for Vl and {v2j} a basis for V2, then the set of tensors
{^i, v2j} is a basis for V1 (g> V2. Thus the dimension of the tensor product
Vl V2 is the product of the dimensions of V^ and V2 , and any tensor can be
written as
40 Lie Groups and Lie Algebras 1.19
for all vl'mVl and all v2 in V2 . The process of replacing the bilinear mapping ft
by the linear mapping a is called lifting, and we say that ft is lifted to a.
Some special cases of the process of lifting bilinear mappings to linear
mappings should be mentioned. Any pair of linear mappings between vector
space determines a linear mapping between the corresponding tensor
product spaces. If a t : V -> V\ and a2 : V2 - V2 are linear mappings, then we
first obtain a bilinear mapping from Vt x V2 to V\ (g) V2 which takes the
ordered pair (i^, v2) into the tensor a^) (g) a 2 (u 2 )- This bilinear mapping
can then be lifted to a linear mapping denoted by a t a2 called the tensor
product of the linear mappings tx.1 and a 2 . The linear mapping o^ a 2 :
Vl (g) V2 -* Ki (g) F2 thus obtained has the fundamental property that
If F is a vector space over the scalar field F, then the mapping which takes
(a, v) into au is a bilinear mapping F x V -* V, which can be lifted to a linear
mapping F (g) V -> V. This linear map is an isomorphism which may be used
1.19 Tensor Product of Vector Spaces 41
is called a Lie algebra. If e ^ , , en is a basis for a real Lie algebra L, then the
Lie product is determined by the structure constants cy* defined by
for all A, fj. e C and all x e L. Finally, the complex vector space C L may
be regarded as a complex Lie algebra if we set
The direct sum, like the tensor product, is a fundamental vector space
operation which finds many applications in the theory of Lie algebras and
their representations. The direct sum operation in vector space theory is
useful both as an analytical tool and as a constructive procedure. Corre-
sponding to these two modes of usage, there are actually two slightly different
definitions of the direct sum, known as the internal and the external direct
sum. In practice there is little danger in being a bit careless on this point since
these two variants are to a large extent equivalent, and the distinction between
them can usually be understood from context. If Vl and V2 are vector spaces
over a field F, their external direct sum consists of all the ordered pairs
(vl, v2), where v: is in Vl and v2 is in V2. The external direct sum, denoted by
Vl + V2, may be regarded as a vector space if vector addition and multiplica-
tion of vectors by scalars are defined componentwise. A familiar example of
the use of the external direct sum is in the construction of the n-dimensional
vector space
over a field F. Another example of the use of the external direct sum is in the
construction of the tensor algebra, to be discussed later.
The concept of internal direct sum is used when we are talking about
the lattice of all subspaces of a given vector space. The set of all subspaces of
a vector space is closed under the operation of intersection, but not under the
operation of union. The sum S: + S2 of subspaces Si and S2 of a vector space
is the set of all elements of the form x + y, where x is in S x , and y is in S 2 .
Equivalently, the sum of two subspaces may be described as the subspace
spanned by their union. The set of all subspaces of a vector space is said to be
a lattice under the two operations D and +. This means that these operations
satisfy certain axioms somewhat reminiscent of Boolean algebra, but not
quite as strong [112], [134]. In particular, the distributive laws between fl
and + do not hold, and there is no analogue of the de Morgan laws of com-
plementation. More formally, a lattice is a partially ordered set in which
1.21 The Lattice of Ideals of a Lie Algebra 43
every pair of elements has a least upper bound and a greatest lower bound.
In the case of the lattice of subspaces of a vector space, the partial ordering is
just the inclusion relation, while S t fl S2 is the greatest lower bound of Sl
and S2 and Sl + 52 is the least upper bound of S t and S2.
The role of complementation in set theory is replaced in vector space
theory by the concept of supplement, which is defined in terms of the internal
direct sum as follows. The sum of two subspaces is said to be an internal
direct sum when the intersection of the two subspaces consists of the zero
vector. If Vis the internal direct sum of subspaces S and T, written V S T,
then every element in V can be written in a unique fashion as the sum of a
vector in S and a vector in T. In this case we say that the subspace T is a
supplement of the subspace S, and conversely. Given any subspace S of a
vector space V, there always exists a subspace Tsuch that V = S 7", but T
is not unique. Thus every subspace has a supplement, but unlike the situation
regarding complementation in set theory, supplements are not uniquely
determined.
Since the two definitions of direct sum are different, their properties
may also be expected to be slightly different. Note that the internal direct sum
is only defined for subspaces of a given vector space having zero intersection.
The external direct sum on the other hand is defined for an arbitrary pair of
vector spaces. Both and 4- satisfy commutative and associative laws, but
they are slightly different in detail. For example, although Sl S2 is equal
to S2 Si, the spaces V + V2 and V2 + V{ are in general not equal, but only
isomorphic. For vector spaces, a homomorphism is any linear mapping, while
an isomorphism is a one-to-one onto linear mapping. If Fis the internal direct
sum of two subspaces Sj and S2, then V is isomorphic to the external direct
sum of S1 and S2. Two vector spaces are isomorphic if and only if their
dimensions are equal. Therefore, the isomorphism between internal and
external direct sums implies that the dimension of the internal direct sum of
two subspaces equals the sum of their dimensions.
The study of Lie algebras in general can be reduced to the study of two
special classes of Lie algebras: solvable Lie algebras and semisimple Lie
algebras. These types of Lie algebras may be defined in terms of ideals as
follows : With any Lie algebra L, there is associated a derived series of ideals
defined recursively by
and
then L' consists of the multiples of e: , and L" = 0, proving that L is solvable.
In any Lie algebra, the solvable ideals form a sublattice of the lattice of
all its ideals, because the sum and intersection of solvable ideals are again
1.23 Semisimple Lie Algebras 45
solvable ideals. In particular, the sum of all the solvable ideals in a Lie
algebra is its unique maximal solvable ideal, called the radical of the Lie
algebra. We may define a semisimple Lie algebra as a Lie algebra which has
no Abelian ideals, other than 0. It is easy to see that a semisimple Lie algebra
cannot have any solvable ideals either, and hence a Lie algebra is semisimple
if and only if its radical is zero.
According to the Levi theorem [151], [245], any Lie algebra can be de-
composed as the direct sum, in the sense of vector spaces, of its radical R
and a semisimple subalgebra S:
That is, each element of the Lie algebra can be written uniquely as the sum
of an element in the radical and an element of this semisimple subalgebra.
Although the radical is uniquely determined, the Levi decomposition is not
unique because there may be several suitable semisimple subalgebras. In fact,
the Levi decomposition L = R S is unique only in the trivial case when S is
an ideal in L. Malcev, Goto and Harish-Chandra showed that if there are
two Levi decompositions, then there is an automorphism of the whole Lie
algebra which carries the one semisimple subalgebra onto the other [104],
[117], [160]. The problem of studying the structure of a Lie algebra is thus
almost reduced to studying semisimple and solvable Lie algebras.
The only question of structure remaining is to characterize the action of
the semisimple subalgebra S on the radical R. To investigate this point, we
need to use the methods of representation theory which we are about to
discuss. Anticipating a little, we can say that this question boils down to
knowing all the modules over a semisimple Lie algebra. Indeed, the condition
[S, R] c: R is equivalent to the statement that we may regard the radical R
as a module over the semisimple Lie algebra S.
To study the structure of semisimple Lie algebras, one uses the fact that
they can be decomposed in terms of simple Lie algebras. A simple Lie algebra
is a non-Abelian Lie algebra which has no proper ideals at all. Every simple
Lie algebra is, as one would expect, semisimple, for the only nonzero ideal is
the whole algebra; if L' = L, this is not solvable, while if L' = 0, then L is
Abelian and consequently not simple. An example of a simple Lie algebra is
the real Lie algebra so(3, R), the familiar three-dimensional vector space R 3
equipped with the usual vector cross product. To see this, we must argue
that there are no proper ideals in the Lie algebra so(3, IR). Such an ideal
would be a proper subspace S of R3, a line or a plane passing through the
origin such that the vector cross product x x y of any vector x in S by a
vector y in R 3 must lie in S. Since the cross product of two vectors is
46 Lie Groups and Lie Algebras 1.23
perpendicular to both and is nonzero unless the two vectors are collinear or
at least one is zero, no line or plane can have this property.
The fact that the rotation group S0(3, R) is simple (has no proper
normal subgroups) almost follows from the simplicity of its Lie algebra.
Since the closure of a normal subgroup is again normal and since the Lie
algebra of a closed normal subgroup of S0(3, R) is an ideal of its Lie algebra,
we only have to rule out discrete and dense normal subgroups. A normal
subgroup of S0(3, R) is dense if its closure is all of S0(3, R). If g is an element
of any normal subgroup S of S0(3, R), then all elements hgh~l which are
conjugate to g must also be in S. But this is equivalent to saying that every
rotation about any axis by an angle equal to that of the rotation g about its
axis must be in S. If g is not the identity, multiplying it by the inverse of a
rotation /ig/z"1 about an axis which has been shifted a small amount, we
obtain an element ghg~1h~l in S which is not the identity, but as close to it
as we wish. In this way one can further argue that any proper normal sub-
group contains a whole neighborhood of the identity element. Since any
neighborhood of the identity generates the whole rotation group, it follows
that SO(3, R) cannot have any proper normal subgroups at all.
Any semisimple Lie algebra S can be written as a direct sum of simple
ideals,
where U and A are operators and U(0) = 1 is the identity. This equation
comes up in quantum mechanics as the Schrodinger equation for the
evolution operator U(t) of a system whose Hamiltonian operator is \(t),
except for a constant factor. Magnus gave a formal solution
for this problem, which converges in some interval about t = 0 if the operators
are finite matrices and A is a continuous function of t (see [159]).
Wichmann and Norman and Wei treated the case
Norman and Wei show that a solution can also be written as a product
These authors also studied the conditions under which the solutions converge
globally, that is, for all t.
In control theory, one frequently wants to know whether a system can be
adequately steered by a given set of controls [139]. In general, the states that
can be reached after a finite length of time from a given initial state under a
suitable class of controls form a certain submanifold of the state space. A sys-
tem is completely controllable if for any pair of states, there is a time interval
and a control on that interval such that the system trajectory, starting in the
one state, and steered by the control, ends at the other. The isotropy subgroup
48 Lie Groups and Lie Algebras 1.24
of a given state consists of those control processes which, when applied to the
system initially in the given state, at the end return the system to that same
state. If a system is completely controllable, the state space itself can be
identified with the coset space of the group of all control processes modulo
the isotropy subgroup of any fixed state [40] . Kucera and Elliott have shown
that a system is completely controllable if it is controllable locally in a certain
sense [85], [86], [149], [150].
The point of departure for much of this work is a pair of formulas of
Baker and Hausdorff which give log(e* ey) and e~xyex in terms of com-
mutators of x and y. The first few terms of the expansions are
and
where A 2 (f) generates a semisimple Lie algebra. The second problem is that
of solving
also subject to the initial condition U^O) = 1. If one can find a solution of
the first problem, then it is shown that the second problem can be reduced to
1.24 The Baker-Campbell-Hausdorff Formula 49
quadratures. The net effect is thus to reduce the case of a general Lie algebra
to the case of a semisimple Lie algebra. The semisimple subalgebra can be
further decomposed as the direct sum of its simple ideals. Thus we may
decompose A2 as
where the A 2j belong to those simple ideals, and hence mutually commute.
The first problem may then be shown to reduce to solving the problems
REPRESENTATION THEORY
One of the main tools in applications of the Lie theory is the concept
of a representation. A representation of a group is a homomorphism of the
group into a group of linear operators on a vector space. In many applica-
tions it is sufficient to treat representations in a fairly loose manner. For
example, one speaks of vectors, tensors, pseudoscalars, spinors and the like
as being various types of geometrical objects, such as directed arrows,
ellipsoids, and so on [39], [208]. Representations of the rotation group crop
up throughout mathematical physics in the form of spherical harmonics and
Legendre functions, multipole expansions and so forth. The actual representa-
tion concept is often held in the background. We must of course make this
intuitive conception more precise, and moreover we want to make our
discussion apply to any Lie group, not just to the rotation group. However,
we shall restrict our attention to analytic homomorphisms, and we assume
that all vector spaces have finite dimension.
We define a finite-dimensional representation of a Lie group to be an
analytic homomorphism of the Lie group into the general linear group of
invertible linear operators on a finite-dimensional vector space. By fixing
a basis in the vector space, we obtain a corresponding homomorphism onto
a matrix group. Since analytic homomorphisms of Lie groups induce
homomorphisms of the corresponding Lie algebras, every finite-dimensional
representation of a Lie group induces a representation of its Lie algebra.
A representation of a Lie algebra L consists of a vector space V and a
homomorphism /from L into the Lie algebra of all linear operators on V. The
term "representation" strictly refers to the pair (K/), but colloquially it is
often used to refer just to the homomorphism / alone. The requirement
that /be a homomorphism means that if x is in L, then/(x) is a linear operator
on K and/(x) depends on x linearly:
51
52 Representation Theory 2.1
The operator ad x behaves just like a first order differential operator in that
it obeys the Leibniz rule for operating on a product,
Although this Leibniz rule is merely another way of writing the Jacobi
identity, this analogy with differential operators is frequently useful in Lie
algebraic computations. We note parenthetically that we may use the
operator ad x to rewrite one of the symbolic Baker-Hausdorff formulas as
where
so that ad is indeed a Lie algebra homomorphism. The pair (L, ad) is then a
representation of the Lie algebra L, called the adjoint representation of the
Lie algebra L.
2.2 Modules over Lie Algebras 53
The importance of the Killing form derives in part from the fact that a
Lie algebra is semisimple if and only if its Killing form is nonsingular.
Solvability can also be characterized in a simple way in terms of the Killing
form. A Lie algebra L is solvable if and only if the Killing form is zero on the
derived algebra L' = [L, L] , that is, if (x, y) = 0 for all x and y in L'.
A linear Lie algebra is a Lie algebra whose elements are linear operators
in some vector space, and whose Lie product is the commutator. A representa-
tion is faithful if it is a one-to-one mapping. If a Lie algebra has a faithful
representation, then the Lie algebra is isomorphic to the image of the
representation, which is a linear Lie algebra. For a semisimple Lie algebra,
the adjoint representation is faithful. For if x belongs to the kernel of ad, then
for any y in L we have
for all Xj and x2 in L and all v in M. As usual, we can feel free to replace the
bilinear mapping L x M -* M by an equivalent linear mapping L M - M,
if we prefer to do so.
The study of modules over an algebra is really the same as the study of
the representations, the only difference being a point of emphasis. If M is a
module over a Lie algebra L, then for any x in L, let/(x) be the linear operator
on M which maps v in M into the vector xv in M. Then (M,/) is a representa-
tion of the Lie algebra L. Conversely, given a representation (M, /) we can
make the vector space M into a module by defining the product xv of x in L
and v in M to be/(x)y. Thus the concepts of module and representation are
equivalent. In talking about modules we emphasize the vector spaces in-
volved, whereas in talking about representations, we emphasize the homo-
morphisms. Generally speaking, the module formulation is the easier to work
with.
Representations and modules can be defined for Lie groups as well as
for Lie algebras. An analytic representation of a Lie group G is a pair con-
sisting of a vector space M and an analytic homomorphism/of G into the Lie
group GL(M) of invertible linear operators on M. If n is the dimension of M,
then GL(M) is isomorphic to the Lie group GL(n, C) of nonsingular n x. n
matrices over C. Hence, a representation of a Lie group corresponds to a
matrix representation, that is, a homomorphism of the Lie group into the
matrix group GL(n, C).
We may also define a module M over an associative algebra A in much
the same way that we defined a module over a Lie algebra. The main dif-
ference is that the requirement
for all x and y in A and for all v in M. If the associative algebra A has a unity
element e, then we may also want to demand that
Lie modules have many applications within Lie algebra theory. For
example, the adjoint representation is equivalent to regarding any Lie
algebra as a module over itself, the module product being taken to be the
Lie product itself. By a slight generalization of this idea, we may also regard
any ideal of a Lie algebra L as a module over L. Again, any extension of a Lie
algebra L, that is, any Lie algebra containing L as a subalgebra, may be
regarded as a module over L. We have already seen that modules enter into
the structure of Lie algebras via the Levi decomposition. In this case the
radical is a module over the semisimple subalgebra, and the module product
is again the Lie product itself. Lie modules also are useful in the further study
of the structure of semisimple Lie algebras via the theory of roots. In this
case the whole Lie algebra is regarded as a module over a certain Abelian
subalgebra, the Cartan subalgebra. We shall come back to this in 2.7.
Any representation (M,/) of a Lie algebra L can be replaced by a
matrix representation. If a basis elt , enis selected in the vector space M,
then the linear operators /(x) correspond to matrices [/'/x)] defined by
The mapping which assigns the matrix [/'/*)] to x is called a matrix rep-
resentation of the Lie algebra L. If a different basis is selected for M, of
course, the matrices of the matrix representation will all be subjected to a
common similarity transformation. Matrix representations related by a
common similarity transformation are considered to be equivalent or
isomorphic. Equivalence of Lie group representations is defined just as for
Lie algebras. Two matrix representations of a Lie group are equivalent or
isomorphic if they are related by a common similarity transformation.
A homomorphism h of one module M over a Lie algebra L into another
module N over L is a linear transformation h : M -> N which preserves the
multiplication by elements of L, that is,
There are several useful ways of combining modules to obtain new ones,
the most immediate of these being the direct sum. Since modules over Lie
algebras are vector spaces, it is meaningful to talk about their internal and
external direct sums as vector spaces. If M : and M2 are modules over a
Lie algebra L, their vector space external direct sum MJ + M2 can also be
regarded as a module over L by defining the action of L componentwise.
Explicitly, the action of any element x of the Lie algebra L on any vector
(v1, v2) in the external direct sum module Ml 4- M2 is given by
We can pass from modules over Lie groups to modules over Lie algebras by
differentiation, or vice versa by exponentiation. To motivate this discussion
we shall restrict our attention to linear Lie groups and algebras, that is,
groups and algebras whose elements are linear operators in a finite-dimen-
sional vector space. If x is an element of a linear Lie algebra L, then g = etx
= 1 4- tx + is an element of a linear Lie group. For sufficiently small real
numbers t, we could ignore higher order terms. Then, inserting this expan-
sion for g in the above definition of its action on the tensor product space,
and comparing powers of f, we arrive at the formula
We may now drop the restriction to linear Lie algebras by using this
formula to define tensor products of modules over abstract Lie algebras. If
M j and M 2 are modules over a Lie algebra L, this formula defines an action
of L on the tensor product vector space Ml M2 making it into a module
over L. It can be verified directly that the axioms for a module over a Lie
algebra are satisfied without reference to Lie groups. As a mnemonic aid
we may further note that the rule
Thus, here DJ,-(x) is the matrix representative associated with the module M,
and the negative transpose D'/x) gives the corresponding matrix repre-
sentation associated with the dual module M*.
with
This suggests the idea of forming the contravariant tensor algebra T(V)
on a vector space Kover a field F. The tensor algebra is defined as the infinite
weak external direct sum
of all the tensor powers of the vector space. The infinite sum here is weak
in the sense that we consider only finite linear combinations of the elements
in the spaces F, V, V V, , so that no actual infinite summations ever
occur, and no questions of convergence can arise. Being a direct sum of
vector spaces, the tensor algebra is clearly a vector space, and with the
tensoring together operation as multiplication, T(V) is an associative
algebra with unity. If V is an n-dimensional vector space, then each tensor
power (X)r Fhas dimension nr, and hence the tensor algebra T(V) is infinite-
dimensional.
Another algebra, closely related to the tensor algebra, is obtained by
antisymmetrizing the tensor product. The resulting theory of the exterior
algebra provides an elegant way to treat determinants, and has important
applications to the theory of Lie groups and Lie algebras. The antisym-
metrization process can be effected most quickly by means of a quotient
construction as follows.
A two-sided ideal I in an associative algebra A is a subspace which
satisfies IA e / and AI c /. The quotient algebra of an associative algebra A
with respect to a two-sided ideal / is the quotient vector space under the
multiplication
is an associative algebra with unity called the exterior algebra or the Grass-
mann algebra of the vector space V. It is customary to use the symbol A
to denote the multiplication operation induced by (g) in the exterior algebra
E(V), so that we have
2.5 Tensor and Exterior Algebras 61
for the cosets of any tensors tl,t2e T(V). If S t and 52 are any two sub-
spaces of the exterior algebra E(V), then S1 A S2 denotes the subspace of
all linear combinations of elements s1 A s 2 , where slSl and s 2 eS 2 .
Since E(V) is an associative algebra, we have
for any three subspaces of the exterior algebra. The vector space Kis naturally
embedded in the exterior algebra E(V), just as it is in the tensor algebra, and
we may identify the vector x e V with the coset x + A e E(V}.
The exterior power
(r copies)
is a subspace of the exterior algebra. Elements of the exterior power /\rV
are sometimes called r-vectors. The whole exterior algebra E(V) is the direct
sum of the exterior powers AT Thus the exterior algebra can be broken up
into pieces each with its own degree, and it is also a graded algebra. The
grading of the exterior algebra is directly related to the corresponding
grading of the tensor algebra. The ideal A is the direct sum of the individual
pieces Ar A fl ((X)T) lying in each tensor power, and the quotient space
((X)T)//T is naturally isomorphic to the exterior power AT. The exterior
square V A V of a vector space V is a vector space generated by exterior
products x A y of vectors x, y in V. If x V, then
we find that
where x and y are in V. More generally, the exterior power f\kV may be
identified with the subspace of (x)fcK consisting of all linear combinations
of elements of the form
where the x,- are vectors in V, and the sum goes over all permutations n of
the indices 1, , k. The sign factor <5n here depends on the parity of the
permutation. It is +1 for even permutations and 1 for odd permutations.
If a is a linear mapping from a vector space Vl into a vector space V2 , then
is, therefore, also a module over the Lie algebra L. The action of an element
x of L on the elements of ArM is induced by the action of x in (X)rM, namely,
where x and y are elements of L. The ideal K thus contains the differences
between Lie algebra products and the corresponding commutators in the
associative tensor algebra. If we consider the associative quotient algebra
64 Representation Theory 2.6
U(L) = T(L)/K, then Lie products will not be distinguished from, com-
mutators since they belong to the same coset. The associative algebra
is called the universal enveloping algebra of the Lie algebra L. As with any
associative algebra, we can also make U(L) a Lie algebra using the com-
mutator operation as the Lie product. If we do this, we can consider L to be
injected homomorphically into U(L), considered as a Lie algebra.
The associative algebra U(L) is useful because of the following property.
Suppose that A is an arbitrary associative algebra, and that A is also given
the commutator Lie algebra structure. Any homomorphism of L into A,
considered as a Lie algebra, has a unique extension to an associative algebra
homomorphism of U(L) into A. Now a representation of a Lie algebra L is
a Lie homomorphism of L into the associative algebra of all linear operators
on the module, with the Lie product of two linear operators being their
commutator. Thus every representation of a Lie algebra L can be extended
to a representation of its universal enveloping associative algebra U(L), and
we see that every module over L can also be regarded as a module over its
enveloping algebra U(L). This idea is central to certain proofs of complete
reducibility for modules over semisimple Lie algebras which are based on
the universal enveloping algebra.
The associative algebra U(L) acts on the module M by letting
where we take the trivial monomial 1 in case n = 0, span T(L), and hence
their cosets span U(L). A result of Poincare, Birkhoff and Witt is that if we
only take monomials having their indices ij in ascending order, allowing
repetition, then the cosets of these monomials, again including 1, form a
basis for U(L).
It can be shown that U(L) has no nonzero zero divisors and that, for
finite-dimensional L, there cannot be an infinite sequence of ideals or one-
sided ideals in U(L) each properly contained in the next. If L is finite-
dimensional, then every ideal of L7(L) is finitely generated. From this, one
can conclude by the Goldie-Ore theorem that U(L) has a (two-sided) divi-
sion ring of fractions [61], [98], [99], [135]. A division ring D is said to be a
division ring of fractions for U(L) if U(L) is a subring of D and every element
of D can be written as a~ 1b and cd~ l, where a, b, c and d are elements of U(L).
2.7 Nilpotent and Cart an Subalgebras 65
Lie algebras over the complex number field we can develop a rather similar
albeit less detailed theory. To obtain direct sum decompositions of modules
over nilpotent Lie algebras, it is similarly necessary to generalize somewhat
the concept of a simultaneous eigenvector.
A nonzero vector v in M is called a weight vector, or generalized simulta-
neous eigenvector, if there exists an integer p such that
for all x in L. The set of all weight vectors corresponding to a given weight u,
together with the zero vector, is a submodule M of the module M, called
the weight submodule for the weight p. In the weight submodule M each
element x in L is represented by an operator which differs from a multiple
of the unit operator by a nilpotent operator. For a module over a nilpotent
Lie algebra we can find a basis consisting entirely of weight vectors. In other
words, a module over a nilpotent Lie algebra is the direct sum of its weight
submodules.
To apply these ideas to a semisimple complex Lie algebra L, we make
use of one of its Cartan subalgebras H. Any module M over the semisimple
Lie algebra L is also a module over the Abelian subalgebra H. Any module
over a semisimple Lie algebra L can then be written as the direct sum of its
weight submodules with respect to H :
for all h e H. Roots play the same role for the structure of Lie algebras that
weights play for Lie modules. The concept of root is in fact just a special
case of the concept of weight which applies to the adjoint representation of
the Lie algebra. Thus, regarding L itself as a module over H via the adjoint
representation, we obtain a decomposition of L as a direct sum of its root
spaces LaH. The Cartan subalgebra H itself is just the root space for a = 0
since H = L?H.
For semisimple Lie algebras over the complex numbers, the above
results may be further sharpened. The restriction of the bilinear Killing form
68 Representation Theory 2.9
for all h e H, and thus we can identify the dual space H* with H itself. The
root system has a number of special properties for semisimple Lie algebras.
For example, if a is a nonzero root, so is a, but there are no roots which
are multiples of a other than a, 0 and - a. Moreover, for semisimple complex
Lie algebras, the root spaces LaH corresponding to nonzero roots a are all
one-dimensional [81]. Thus, for any nonzero root a, there is, up to a factor,
a unique vector ea e L, called a root vector, such that
for all h e H. In other words, for all h in H, this root vector ea is a simultaneous
eigenvector of the linear operators ad h acting on L.
If ea and ef are root vectors, then by the Jacobi identity,
Thus, a basis for H together with the ea determines a basis for L. It follows,
therefore, that the total number of nonzero roots of a semisimple Lie algebra
is equal to its dimension minus its rank. If a and /? are roots such that
a + ft 96 0, then the corresponding root vectors ea and e^ are orthogonal.
Hence ea and e_ a are both null vectors, but they are never orthogonal to
each other, and hence the bilinear Killing form restricted to the plane
spanned by them is nonsingular.
To further study the root system of a complex semisimple Lie algebra,
we introduce a certain Euclidean space and an ordering of the roots. Let
H% denote the real /-dimensional subspace of H* whose elements are all
real linear combinations of roots. Since H* can be identified with H, we
may define a metric on the dual space H* by
2.9 Roots of Semisimple Lie Algebras 69
for all a and ft in H *. If a and ft are nonzero roots, then (a, ft) is real and (a, a)
is positive nonzero. The Killing form is thus positive definite on the real
space H|, making this a Euclidean space with the Killing form as inner
product. We obtain a total ordering of the roots by choosing an arbitrary
ordered basis y l 9 , yl for the real Euclidean space H.
Thus, /* + v is one of the weights in the tensor product of these weight sub-
modules, and in fact it is the only weight in this tensor product.
In general, the weight system of the tensor product M N of any two
modules M and N is the collection of sums of a weight of M and a weight
of N. That is, the weight system of a tensor product of two modules can be
obtained from the weight systems of the two individual modules by a process
of vector addition. These results can be used as follows to clarify the action
of a Lie algebra L on one of its modules M. This action L x M -> M can
70 Representation Theory 2.1 0
The value of the concepts of roots and weights is not limited to semi-
simple Lie algebras. One example of the use of these concepts in a more
general setting arises in the theory of the special functions of mathematical
physics. In addition to the elementary transcendental functions, such as ex
and sinx, these special functions include the Bessel functions, Legendre
polynomials, parabolic cylinder functions, Hermite polynomials and so forth.
Many of them share a variety of interesting properties, such as recurrence
relations, addition theorems, generating functions, orthogonality relations,
integral representations and so on. A partial explanation of these regularities
lies in the fact that many of these functions may be regarded as special cases
of the ordinary or confluent hypergeometric functions. A deeper reason,
perhaps, is that many of these functions arise from the solution of the Laplace
equation upon separating variables. In general, special functions arising from
the solution of the Laplace equation or its generalizations are called harmonic
functions or spherical functions [102], [162], [232]. Now, the Laplace equa-
tion is an example of a differential equation invariant under the action of
a certain Lie group. Hence, it is reasonable to suppose that the theory of
special functions can be unified and also generalized by making use of ideas
2.10 The Factorization Method and Special Functions 71
from the theory of Lie groups and Lie algebras [225] , [234] . An obvious
generalization, for example, would be to replace the Laplace operator for
flat Euclidean space R" by an analogous operator for a curved symmetric
space such as a sphere [243] .
One important method for studying special functions via their recur-
sion relations, known as the factorization method, ties in closely with the
standard Lie algebraic techniques involving roots and weights. Infield and
Hull gave a systematic catalogue of the special functions which can be studied
by this method, which involves factorizing the appropriate second order
differential equation [128]. For the Hermite polynomials, which are related
to the harmonic oscillator, this leads to the creation and annihilation oper-
ators considered previously. Hull and Infield reduced the entire list of
possibilities to six overlapping classes, later shown by Miller to arise naturally
in the representation theory of four particular Lie algebras [168], [169].
These include the Lie algebras of the three-dimensional rotation group and
the groups of Euclidean motions in a plane and in space, as well as a certain
four-dimensional solvable Lie algebra. The Lie algebras that one studies are
those generated by the differential operators which figure in the recursion
relations. The recursion relations for the special functions themselves can
be viewed as examples of the shifting rule for root vectors acting on weight
vectors.
As an example, we discuss the Bessel functions, which are related to the
three-dimensional Lie algebra of the group of Euclidean motions in a plane
[233], [248]. Any Euclidean motion in the plane can be composed out of a
translation and a rotation about the origin. The angle </> of the rotation and
the orthogonal components (x, y) of the translation may serve as local
coordinates for a neighborhood of the identity in the Euclidean group of a
plane. We denote by Tx, Ty and M tangent vectors corresponding to the
one-parameter subgroups consisting, respectively, of translations in the x
direction and y direction, and rotations about the origin. We can identify
the basis vectors Tx, Ty and M for the Lie algebra with differential operators :
This yields an isomorphism of the Lie algebra of the plane Euclidean group
with a Lie algebra of differential operators on the infinitely differentiable
functions on the plane, the Lie product being commutation. Note that the
element
by using polar coordinates (r, <j>) in the (x, ,y)-plane. In fact, it is easy to
transform this equation into the standard differential equation for the Bessel
function Jn(r). The complexification of the real Lie algebra of the plane
Euclidean group is not semisimple since the subspace spanned by Tx and
1y is an Abelian ideal. This Lie algebra is in fact an example of a solvable
Lie algebra which is not nilpotent.
The line spanned by M and the plane spanned by Tx and Ty are both
maximal Abelian subalgebras. However, only the line spanned by M is a
Cartan subalgebra, while the plane spanned by Tx and Ty is not. By setting
Since M^m = imij/m, the vector \j/m is an eigenvector of M, and hence spans
a weight submodule. The elements T + and T_ act as raising and lowering
elements, satisfying the shifting rule
The shifting rule then reduces to the usual recursion relation for Bessel
functions,
2.11 The Cartan Matrix 73
is called an en-ladder through ft if they are all roots and if ft (p + l)a and
ft + (q + l)a are not roots. Then p and q are related by
The vectors e l 5 , et, called simple raising elements, and the vectors
/! , , /, , called simple lowering elements, together generate the whole Lie
algebra. The Lie products are given by
We caution the reader that in these formulas one does not sum over the
repeated indices.
The relations written above do not yet completely define all products,
for (X, ej\ and [/,/,-] have not been specified. The complete set of relations
can nevertheless be derived from the above set by a well-defined algorithm
which we shall shortly illustrate in a special case. The elements hit together
with
and
span the Lie algebra. All linear relations and commutation relations for
these elements are computable from the Cartan matrix.
The roots of a semisimple Lie algebra and the weights of any of its
modules are vectors lying in the dual space H* of a Cartan subalgebra H.
Moreover, these weights, just as the roots, are rational linear combinations of
the simple roots o^, , a,. Hence the weight system of any module over a
semisimple Lie algebra of rank / is a set of points lying in the /-dimensional
real Euclidean space H% of all real linear combinations of the simple roots.
Since this Euclidean space is lexicographically ordered with respect to an
arbitrary basis, we may regard the weights of a module as a totally ordered
set. For a finite-dimensional module, there are only a finite number of dif-
ferent weights, and hence there must be a highest weight which is higher than
all the others. Two irreducible modules over a semisimple Lie algebra are
isomorphic if and only if their highest weights are equal. Thus the irreducible
modules over a semisimple Lie algebra are characterized by their highest
weights.
To study the structure of an irreducible module with a given highest
weight A, we must have a way of calculating its weight system. The weight
2.12 The Weyl Group 75
The weight system of any module is invariant under the Weyl group.
The Killing form also is invariant under the Weyl group. If w is any element
of the Weyl group W, then
In general, the root system may have a higher degree of symmetry than
some of the other weight systems. This is because the root system is not only
invariant under the Weyl group, but also under the inversion a H* a.
The sets of vectors which can represent root systems of semisimple Lie
algebras are restricted by the following three conditions :
1. If a is a root, then a is a root, but no other nonzero multiples of a
are roots.
2. If a and /? are nonzero roots, then
isO, 1, 2 or 3.
3. The Weyl reflection of a root jS in the hyperplane through the origin
perpendicular to any nonzero root a yields another root,
From these three properties of the root system, one can deduce a number
of other simple properties. For example, it follows that the angle between
any two nonzero roots can only be 0, 30, 45, 60, 90, 120, 135, 150 or
180. Also, the relative lengths of any two nonzero and nonorthogonal roots
are determined by the angle between them. If this angle is 0, 60, 120 or 180,
then the roots are of equal length. If the angle is 45 or 135, their relative
lengths are in the ratio ^/2:l, while if the angle is 30 or 150, then this
ratio is >/3:l. In fact, these three conditions alone determine the possible
root systems, and hence all the possible semisimple Lie algebras over the
complex numbers.
76 Representation Theory 2.13
While the problem of finding all semisimple complex Lie algebras has
thus been reduced to a geometrical problem, Dynkin found an even simpler
way of doing this by using diagrams in a plane [81] . He represents a system of
simple roots al , , a, by a set of / points in a plane, where / is the rank of the
Lie algebra. These points form the vertices of the Dynkin diagram. The
points representing a, and a,- are connected by
lines. The number of lines determines the angle between the roots, since
(a,, a,-) ^ 0 for simple roots. Thus the angle 4-a,- / is respectively 90, 120,
135 or 150 if the number of lines which connect the /th and jth vertices is
zero, one, two or three. While the Dynkin diagram of a simple Lie algebra
is connected, the Dynkin diagram of a semisimple Lie algebra which is not
simple is not connected. The connected components of the Dynkin diagram
of a semisimple Lie algebra correspond to the simple ideals occurring in a
direct sum decomposition of the Lie algebra.
The relative lengths of the simple roots are indicated in the Dynkin
diagram by using light and dark circles to represent the vertices. If all roots
are of the same length, all the circles are light. For simple Lie algebras, the
only other possibility is that there are roots of two different lengths. The long
roots are then assigned light circles, the short roots dark circles. The Dynkin
diagram thus allows one to compute the angles between the simple roots and
their relative lengths. The absolute lengths of the roots may be obtained from
an observation of Brown that the sum of the squares of the lengths of all the
roots of a semisimple Lie algebra is equal to its rank [41]. From this informa-
tion one can then draw the root system as a set of points in /-dimensional
Euclidean space. The ratios of the square of the lengths of the simple roots are
related to the Cartan matrix by
Since the product AijAy and the ratio Atj/Ajt can be determined from the
Dynkin diagrams, we can obviously calculate all the Aijt at least up to signs.
The sign is always minus for nonzero off-diagonal elements, and the diagonal
elements are clearly Ati = +2. Thus the Dynkin diagram determines the
Cartan matrix uniquely. From a knowledge of the Dynkin diagram one can
therefore reconstruct the Lie algebra, so that the Dynkin diagram completely
characterizes a given semisimple Lie algebra.
The three conditions which characterize root systems can be reinter-
preted as various restrictions on the possible Dynkin diagrams of simple
2.13 Dynkin Diagrams 77
Lie algebras. For example, a Dynkin diagram cannot contain any loops, and
the total number of lines incident on a given vertex can be at most three.
Continuing in this fashion, we can narrow down the possible connected
Dynkin diagrams to four general classes and five exceptions. Dynkin dia-
grams of the four main classes of simple Lie algebras are shown in Fig. 1,
while those of the five exceptional simple Lie algebras are shown in Fig. 2.
FIG. 1. The four main classes of simple Lie FIG. 2. The five exceptional simple Lie
algebras algebras
The dimensions of the main-series complex simple Lie algebras are given by
2.15 COnstruction of the Lie Algebra A2 79
The two simple roots o^ and a2 are of equal length, and form an angle of
120. Hence the Cartan matrix, with components Atj, of the Lie algebra A2 is
Since the difference of two simple roots obviously cannot be a root, we have
p = 0 for the a Bladder through <x 2 , and hence,
and this ladder is thus composed of the roots a2 and ct1 + a 2 . By considering
further ladders, we obtain no more positive roots, and we see that o^ + a 2
is the highest root. Thus the root system of the simple Lie algebra A2 consists
of zero, the positive roots a x , a 2 and a t + a 2 and the negative roots
a l s a2 and (a.1 + a 2 ). The Euclidean space H% spanned by the simple
roots <*! and 2 is a plane, and thus we can draw the root diagram of the Lie
algebra A2, as shown in Fig. 4.
From this figure it is seen that the root system of A2 is not only symmetric
under the Weyl group, but also possesses an additional symmetry under
inversion in the origin.
We obtain part of a canonical basis for A2 by considering the elements
h1,h2 and the simple raising and lowering elements e , e2 and /i ,/2 . Their
commutation relations are
and, of course,
which are root vectors corresponding to the roots (a1 + a2), to obtain a
total of eight vectors which form a basis for A2.
The commutation relations involving the elements el2 and /12 are
obtained by using the Jacobi identity, for example,
We recall that there are four general series of simple Lie algebras over
the complex numbers, At,Bt,Ci and D,, and five exceptional simple Lie
algebras, G 2 , F4, E6, 7 and 8. Of course, for / < 4, some of the general
sequence Lie algebras are not defined or they collapse into one another, so
we have only A1, A2, B2, A3, B3 and C 3 . Real simple Lie algebras, such as
2.16 Complexification and Real Forms 81
the algebra so(3, R), are also useful for many applications. The prob-
lem of classifying the real simple Lie algebras is only slightly more
complicated than classifying the complex ones. This is because the process of
complexification relates the real case to the complex case. Hence, we may
study the real simple Lie algebras by generating them from the complex
algebras. Since a single complex simple Lie algebra in general yields several
real ones, we must supplement our study of Dynkin diagrams with further
considerations [93], [94].
We recall that the complexification of a real Lie algebra is obtained by
tensoring it with the complex number field C regarded as a vector space over
the real numbers. Conversely, we obtain the real restriction LR of a complex
Lie algebra L by simply regarding L itself as an algebra over the real subfield
of the complex number field. Note that the processes of complexification and
real restriction are not inverses of each other. The complexification of the
real restriction of a complex Lie algebra Ldoes not yield back L, but rather
C LR L 4- L.
If we can write the real restriction LR of a complex Lie algebra L as a
direct sum L0 i'L0 for some real subalgebra L0, then we call L0 a real
form of L. If L0 is a real form of a complex Lie algebra L, then conversely,
L is isomorphic to the complexification of L0.
Every real simple Lie algebra is either the real restriction or else a real
form of a complex simple Lie algebra. The complication of the classification
problem for real simple Lie algebras arises from the fact that a complex simple
Lie algebra may have several distinct real forms. Or, putting it another way,
several different real simple Lie algebras may have isomorphic complexifica-
tions. Consider for instance the real Lie algebra so(3, R) of the rotation
group and the algebra so(2,1; R) of the plane Lorentz group, that is, the
group of linear transformations on x, y and t leaving x2 + y2 t2 invariant.
These real Lie algebras are not isomorphic, but they have isomorphic
complexifications, namely the complex simple Lie algebra Al.
Of all the real forms of a given complex simple Lie algebra, there is
precisely one, called the compact real form, which is the real Lie algebra of
a compact Lie group. In general, a compact real Lie algebra is defined to be
a real Lie algebra obtained from some compact Lie group. A useful character-
ization of compact semisimple Lie algebras arises from the fact that a real
semisimple Lie algebra is compact if and only if its bilinear Killing form is
negative definite. All noncompact real forms of a simple complex Lie algebra
can be obtained from the compact one [178], [206]. A detailed list of the real
simple Lie algebras, together with their isomorphisms, was found by Cartan
[46], [120].
Any real form L0 of a complex Lie algebra L may of course be regarded
as a subalgebra of the real restriction LR. If et, , en are a basis over the
real numbers for the real form L 0 , then these same vectors are a basis for
L over the complex numbers. Moreover, the vectors ev, , en, ie^, , ien
are linearly independent over the real numbers, and are a basis for the real Lie
82 Representation Theory 2.16
where x and y are elements of L0. We may define an automorphism 5" of the
real Lie algebra L* by
The vectors ihlt , ihh together with i(ea + e_J and ea e_x for all
a > 0, then form a basis over the real numbers for the compact real form L0
of the complex simple Lie algebra L. This particular choice of basis makes it
easy to show that the Killing form is negative definite on the compact real
form L0.
Any involutive automorphism 50 of the compact real form L0 of a
complex simple Lie algebra L can be extended to an automorphism of the real
Lie algebra
by defining
of the compact real form L0 with respect to S0 . We then find that the vectors of
2.17 Real Forms of the Lie Algebra >41 83
are left fixed by S, while those of iL^ are carried to their negatives. To show
that S is an involutive automorphism of
for all x and y in L0 . Since L1 consists of all vectors left fixed by S, we see that
if x and y are in Ll , then so is [x, y] , for we have
The complexification of s(2) then has the same basis and same Lie products
over the complex numbers, this complex Lie algebra being, of course, the
complex simple Lie algebra A^. For the benefit of physicists we remark that
it is really this complex Lie algebra Al which is usually studied in the quan-
tum theory of angular momentum. In angular momentum theory it is
conventional not to use the basis elements el,e2,e3, but to replace them
with an obviously equivalent set
To discuss the real forms of the complex Lie algebra Alt we use still
another basis, which corresponds to the root space decomposition of this
algebra. The Cartan subalgebras of the complex simple Lie algebra At are
one-dimensional. The selection of a Cartan subalgebra is described in
angular momentum theory as choosing a quantization axis. It is conventional
to choose h = ;3 = ie3 as the basis of a Cartan subalgebra, but of course
one could equally well have chosen jl or j2 for instance. Note however that
84 Representation Theory 2.1 7
The roots of the Lie algebra A x with respect to H are the three linear forms 0,
+ a and a. The corresponding root vectors are
and
which together with h form a basis for the three-dimensional complex Lie
algebra A1 . We have
The compact real form of A^ is the original real Lie algebra su(2) *
so(3, R), since the corresponding Lie groups SU(2) and SO(3, R) are both
compact. The compact real form can also be obtained from the general
prescription given earlier. According to (his prescription, we should have
for all x and y in the Lie algebra If all the eigenvalues of S0 were - 1, then
S0 = 1, and
while
and
follow from the fact that S0 is an automorphism. Since [x,y] is the vector
cross product of x and y, it is a nonzero vector orthogonal to both x and y,
and hence LO must be two-dimensional, containing the two linearly in-
dependent vectors 3; and [x, y], while LQ is one-dimensional. The vectors
3; and [x,y] are an orthogonal basis for the eigenspace LQ, and since
[LQ,LQ] c LQ, we see that
But this implies that [y, [x, y]] is a multiple of x because LQ is one-dimensional,
and hence x is orthogonal also to y. Thus x, y and [x, y] are mutually orthog-
onal vectors, and if x and y are taken to be unit vectors, then [x, y] will
also be a unit vector. We can assume that et = x, e2 = y and e3 = [x, y]
without loss of generality. There is thus, up to isomorphism, only one
noncompact real form of A t , corresponding to the automorphism
The real Lie algebra so obtained is the real Lie algebra of several Lie
groups, including the plane Lorentz group SO(2,1; R) leaving invariant
the form
the real special linear group SL(2, R), the real symplectic group Sp(l, R)
and the pseudo-unitary group 517(1,1). All of these Lie groups are, of course,
locally isomorphic since their Lie algebras are isomorphic. The representa-
tions of these noncompact groups have been studied extensively [17], [32],
[193].
The real Lie algebras related to the complex Lie algebra Al include not
only the two real forms discussed above, but also the real restriction A* of A.
86 Representation Theory 2.18
The real Lie algebra A* is the Lie algebra of several familiar Lie groups,
which are again locally isomorphic. These include the ordinary homogeneous
Lorentz group 50(3,1; R) leaving invariant the form
the complex special linear group SL(2, C) and the complex orthogonal group
50(3, C).
where we have set h 1 for convenience. Adding also the spin S, for which
there is no completely satisfactory classical analogue, we obtain the total
angular momentum
for a single particle. One now postulates that the operators J are also
Hermitian, and satisfy the same commutation relations as L, namely,
2.18 Angular Momentum Theory 87
If \l/ is an eigenvector in Hilbert space such that J3\j/ = mi/f, then J\j/
are either zero or else eigenvectors of J3 with eigenvalues m 1. This
shows that J+ raises the eigenvalue by one unit and J_ lowers the eigenvalue
by one unit ; hence we have the terms "raising" and "lowering operators."
The theory of the Lorentz group is the relativistic generalization of
angular momentum theory. To develop this theory one may proceed
heuristically, starting with the classical description of a relativistic spinning
particle. The equation
shows that it is an ideal. The generators of the Lorentz subgroup satisfy the
commutation relations
2.18 Angular Momentum Theory 89
Although the real Lie algebra X? of the Lorentz group is simple, its
complexification is not simple, but only semisimple, splitting up as A r + Al.
We may note, by the way, that S0(4, R) and S0(3, R) x SO(3, R) have real
Lie algebras isomorphic to sw(2) 4- sw(2), whose complexification also
yields Al + Al. In other words, both Xf and sw(2) + su(2) are real forms of
the same semisimple complex Lie algebra A^ + Al.
As a consequence, the theory of the Lie algebra of the Lorentz group
may be described as a doubled version of angular momentum theory. In
fact, the commutation relations of the Lorentz algebra, which may of course
be rewritten in terms of J and K, acquire a simple form if we set
CONSTRUCTIVE METHODS
Any module over a Lie algebra may also be regarded as a module over
its universal enveloping associative algebra. We may therefore use the
representation theory of associative algebras to clarify the structure of Lie
modules. For example, the universal enveloping algebra plays an important
part in constructing the irreducible modules over a semisimple Lie algebra.
For this purpose we need to take a quick look at the structure of the envelop-
ing algebra of a semisimple Lie algebra L resulting from its root space de-
composition. We choose a Cartan subalgebra H and denote the subspaces
of L spanned respectively by the root vectors of the positive and negative
roots by L+ and L~'. The subspaces L+ and L~ are nilpotent subalgebras of
L generated respectively by the simple raising and lowering elements, and
we can write
91
92 Constructive Methods 3.1
for a > 0. The extreme vectors together with the zero vector form a subspace,
the extreme subspace of the module. If a module is irreducible, its extreme
subspace is one-dimensional, and is spanned by a weight vector v for the
highest weight L Indeed, for any a > 0, the vector ejo would be a weight
vector for the higher weight A + a if it were not zero. In any module M, the
subspace 17 ~x is an irreducible submodule if x is an extreme vector. More-
over, any basis X j , , xn of its extreme subspace yields a decomposition
of the module M as a direct sum of irreducible submodules
If we regard U(L) as a module over itself, the mapping which takes any
element u E U(L) into the vector ux e M may be described as a module
homomorphism from U(L) onto M. The kernel of this module homomor-
phism is a submodule, and hence a left ideal, of U(L) which we call the anni-
hilator of the cyclic module M. The annihilator A does not depend on x, and
may be described as the set of elements a e U(L) such that av = 0 for all u e M.
By the fundamental homomorphism theorem, the module M is isomorphic to
V(L)/A, the element MX in M corresponding to the coset u + A in U(L)/A.
This result, M w U(L)/A, means that every cyclic module may be obtained
as a quotient of U(L) with respect to some left ideal. The problem of finding
all the irreducible representations of U(L), and consequently of L, is equiva-
lent to that of finding all the maximal left ideals of U(L). Conversely, if A
is a left ideal of U(L), then the cosets form a cyclic module U(L)/A over L,
and this module is irreducible if and only if A is a maximal left ideal.
3.2 Dynkin Indices 93
called the Dynkin indices of the weight //. The Dynkin indices of the highest
weight uniquely characterize an irreducible module over a semisimple Lie
algebra up to isomorphism. In general the Dynkin indices of a weight can be
any integers, positive, zero or negative, but the Dynkin indices of a highest
weight are always nonnegative. The Dynkin indices of the simple roots a,-
are given by the elements of the Cartan matrix [Atj] , where
The action of the Weyl group on the weight system of a module is particularly
easy to compute in terms of Dynkin indices. Each basic weight is invariant
under all but one of the simple Weyl reflections since w^- = kj for i ^ j,
while w^ = A, <X; . Hence the Dynkin indices m- of the reflected weight
Wj/2 are related to the indices ml of /i itself by the relation
in terms of a set of basic modules Mlt , M,. This module is the Cartan
composition
where the weight vector x,- in M,-, for the basic highest weight A,-, appears nt
times in the tensor product. Thus the process of Cartan composition reduces
the problem of constructing the irreducible modules of a semisimple Lie
algebra to that of constructing its basic modules.
Dynkin introduced a convenient way of specifying the irreducible
modules over a semisimple Lie algebra [83]. One simply writes the rth
Dynkin index of the highest weight above the ith vertex of the Dynkin
diagram of the Lie algebra. In Fig. 5 this is illustrated for a particular rep-
resentation of the Lie algebra B3, the complexification of the real Lie algebra
ofSO(7,R).
FIG. 5. Dynkin diagram for the ^-dimensional spinor module ofB3, which can be used to define
the exceptional Lie algebra G2
and 8, each module is isomorphic to its own dual [161]. The dual operation
thus plays no particularly important role for these Lie algebras, while for
the remaining algebras, many but not all modules are self-dual. In practice,
the main application of duality in the representation theory of simple Lie
algebras is to Lie algebras of type Al with / ^ 2. For an irreducible module
over A{, the Dynkin indices of the dual module may be obtained by replacing
m,- b y m / + 1 _ .
There is a Cartan subalgebra H with basis h, and roots 0, <x, where <x(/z) = 1.
Since the sum of the squares of the lengths of the roots equals the rank, which
is one, we have (a, a) = |. The raising and lowering algebras U+ and U~
consist, respectively, of all polynomials in ea and in e_ a . There is one basic
weight ^. in the one-dimensional space H%, and since it is a multiple of a
satisfying
it must be 1 = a/2.
In the quantum theory of angular momentum, the basic module over
A! is known as the spinor representation, or the spin-half representation of
96 Constructive Methods 3.3
the rotation group [190]. The weight vector x corresponding to the highest
weight A satisfies
Since the basic weight A goes over into its negative under a Weyl reflection,
the weight system of the basic module consists at least of +A, and in fact
consists of these two weights alone. Since
it follows that y e.gX is nonzero and hence is a weight vector for the
shifted weight A - a = -A. We must have e.j = 0, because if it were
nonzero, we would obtain a weight lower than A, and by Weyl reflection
one higher than A. Hence the spinor space U~x is a two-dimensional complex
vector space spanned by the two basis vectors x and y. The action of A j on the
basic module is given by
Rewriting this action in terms of the basis JiJ2>J3 yields the Pauli
matrices,
The linear mapping which takes jk into %ak for k = 1,2,3 is precisely what is
called the spin-half representation of the Lie algebra Av.
Every irreducible module over Av is isomorphic to a module generated
by letting the lowering algebra U~ act on a tensor product x (g> x.
Each irreducible module is therefore characterized by a single integer n,
the rank of the tensor x x, that is, the number of repetitions of x.
The tensor x x is the highest weight of the module, and since
the highest weight for the module is (w/2)a = ya. In the terminology of
angular momentum theory, the number j = n/2, which can be either an
integer or a half-integer, is called the spin of the module [84]. The irreducible
modules corresponding to representations of the rotation group are just
those having integer spin, while the others are representations only of its
covering group. We now calculate the entire module structure, choosing
3.4 The Casimir Subalgebra 97
the module basis to agree with the usual conventions of the quantum theory
of angular momentum. In the case n = 2, corresponding to spin one, we
find
More generally, for the irreducible module M, with spin j, we can compute
a basis of 2j + 1 vectors um, where
with
(2j factors),
This basis can be chosen so that the action of the Lie algebra is given by
The element
of the universal enveloping algebra, called the second order Casimir element
of L, commutes with every element of the Lie algebra [49]. The second
order Casimir element for a semisimple Lie algebra is analogous to the
Laplace operator in the theory of special functions. We can also give a
formula for y which is highly reminiscent of the factorization method for
the Laplace operator. If h1, , ht is a basis for a Cartan subalgebra H,
then since H is nonsingular, there is another basis h1, , h1 such that
The sum in the second term goes over all the nonzero roots a of L, and ea is
any nonzero root vector in L.
The second order Casimir element is used in the proof that modules
over semisimple Lie algebras are completely reducible [50]. It also has many
other uses in the theory of representations of semisimple Lie algebras. In an
irreducible module M, the operator corresponding to y is a multiple of the
unit operator, y = y(M) 1. If M has highest weight A, then the eigenvalue
y(M) of the operator representing the second order Casimir invariant in M
is given by Weyl's formula
where <5 is half the sum of the positive roots. In the adjoint representation of
a simple Lie algebra L, we have y(L) = 1, and thus the operator correspond-
ing to y is the unit operator itself. For example, if we take the usual angular
momentum basis jt, j2, ;3 for the Lie algebra Av, then the Casimir element y
is
where ze C(L). The linear form T e C(L)* characterizes the modules in the
sense that if modules M and M' have linear forms T and T' which are pro-
portional to each other, then the modules are isomorphic [116].
The simple Lie algebra A2 is the complexification of the real Lie algebra
of the special unitary group SU(3). This particular group has recently been
studied extensively in applications dealing with the strongly interacting
elementary particles [72]. We recall that the six nonzero roots of A2 have
equal lengths, forming the vertices of a regular hexagon, as shown in Fig. 4
( 2.15). Since the sum of the squares of the lengths of all the roots is equal to
the rank, which is two, each nonzero root has length 1/^/3. The Killing form
can now be computed since the angle between the simple roots a t and a 2 is
120. We find that
and
The canonical basis for A2 consists of the simple raising elements e and
e2 , the simple lowering elements /t and /2 , their commutators el2 = [elt e2]
and /12 = [/i,/2] and the elements hv and /i2. The equations defining the
basic weights ^ and A2 can be solved either algebraically or geometrically.
These equations say that Aj is perpendicular to a 2 , and its projection onto
ocj is half the length of a x , as shown in Fig. 6.
FIG. 6. Basic weights of the simple Lie algebra A2 shown superimposed on the root diagram
100 Constructive Methods 3.5
Using the Weyl group, we find that the weight system of the basic module
with highest weight A t is an equilateral triangle consisting of three weights
/d , fi2 and {JL3 , as shown in Fig. 7. For the basic module Ml we choose a basis
*i , x2 , x3 consisting of weight vectors corresponding to the weights H i , n 2 , f * 3
respectively, where
Having chosen the extreme vector x t , clearly we can define xz and x3 by the
equations
where we used
and A//JJ-) = 6^. The shifting rule LM c M^+a and an inspection of the
weight system shows that
3.6 CHARACTERS
if there is an element s in the group such that g2 = sgjS" 1 . Since the character
takes on the same value at conjugate elements of the group, it is determined
by its restriction to such a maximal torus subgroup.
This leads us to define the character of a finite-dimensional module
over a semisimple complex Lie algebra as a complex-valued function on a
Cartan subalgebra. The character % of a module M over a Lie algebra L with
Cartan subalgebra H is defined by
Characters are also useful for computing the Clebsch-Gordan series which
reduce the tensor products of irreducible modules as direct sums of ir-
reducible submodules. The character of the direct sum M + N of two
modules M and N is the sum of their characters :
This may be proved by choosing a basis in the direct sum which is the union
of bases in the summands. The character of the tensor product M N of
two modules is the product of their characters :
into
where the summation goes over all the weights of H in M. This formula
shows that a knowledge of the character of a module is equivalent to knowing
its weight system and the multiplicity of each weight. Practical methods for
computing characters and multiplicities of weights will be discussed later.
We may note that the highest weight of an irreducible module over a semi-
simple Lie algebra has multiplicity one. Also, if w is any element of the Weyl
group, then the weights fj. and w^i have the same multiplicity in any module.
We can write
which can be read off directly from the Dynkin diagram. Since the Dynkin
diagram informs us directly only about relative lengths and angles between
simple roots, we only obtain Mtj and we will later have to compute the
normalization factor N by separate means. We can compute the integer
Cartan matrix from
The formula
(no sum) yields the metric up to the normalization factor D which we can
obtain in two different ways. One way, which requires a prior computation
of the root system, is to use the fact that the sum of the squares of the lengths
of the roots of a semisimple Lie algebra is equal to its rank. For simple Lie
algebras, a better procedure is to use the fact that the second order Casimir
operator has the value 1 in the adjoint representation. If we know the highest
weight of the adjoint representation, that is, the highest root, then this
method is computationally much simpler. Indeed, if rj , , r, are the Dynkin
indices of the highest root, then
and
3.7 Computation of the Killing Form 105
where p is the highest root and <5 is half the sum of the positive roots.
We may illustrate the method for the simple Lie algebra B2 correspond-
ing to the Lie groups SO(5, R) and Sp(2). From the Dynkin diagram shown
in Fig. 8, we obtain
Since the Dynkin indices of the highest root of B2 are (2, 0), we find D = 24
and N = 6, and hence the matrix [g0-] is given by
FIG. 8. Dynkin and Root diagrams of B2, showing also the basic weights
106 Constructive Methods 3.8
The basic inputs for a given weight diagram calculation are the rank
and type of the simple Lie algebra under consideration and the Dynkin
indices of the highest weight of the irreducible module whose weight diagram
is to be determined. The outputs include the weight system, the multiplicity
of each weight, and other pertinent information. Given the highest weight
of an irreducible module over a low rank semisimple Lie algebra, finding all
the other weights is a simple problem in geometry. Since the Weyl group
describes the symmetry properties of weight systems, we need only apply
Weyl reflections to the highest weight to obtain the general shape of the
weight system. We can then fill in the rest of the weight system by using
ladders of weights.
To illustrate this procedure, consider again the simple Lie algebra B2 .
The Weyl group of this algebra is generated by the Weyl reflections w t and
w2 associated with the simple roots a x and a 2 . If we introduce Cartesian
coordinates (x, y) in the plane of the root system, then w^ and w2 are the
following maps :
reflection in the x-axis,
reflection about the line
Note that
split up into layers, the first layer consisting of the highest weight alone [83],
[135]. Each weight in a given layer may be obtained from some parent
weight n contained in the previous layer by subtracting some simple root a,.
To obtain a particular layer, given all the preceding layers, we consider the
arladders passing through all the weights
belonging to the previous layer. Since the top and bottom weights of the
ladder
are related by the simple Weyl reflection wt corresponding to the root a,,
we have
Since the weights lying above // in this ladder belong to previous layers, the
value of q for this ladder is known and we can compute p = q + m . The
weight /j. a,- belongs to the given layer if and only if p is positive.
We illustrate this algorithm in Fig. 10, which summarizes the computa-
tion of the weight system for the 16-dimensional irreducible module over
B2 with highest weight (1, 1).
FIG. 10. Computation of weights for the 16-dimensional irreducible module for B2 with highest
weight
108 Constructive Methods 3.8
From the transpose of the Cartan matrix we can read off the simple
roots, which turn out to be (2, -1) and (-2,2). The weight system, con-
sisting of a total of 12 weights, is shown in Fig. 11.
FIG. 11. Weight diagram (showing multiplicities) for the irreducible module for B2 with highest
weight A A, + <12
where
and B is the matrix related to the inverse of the Cartan matrix. Then /* > 0,
for example, if and only if the first nonzero entry in the new array (p 4 , , p{)
is positive. Thus the process of lexicographic ordering requires a previous
computation of the inverse of the Cartan matrix. The final output of the
Dynkin layer method is a table of the Dynkin indices of all the weights of a
given irreducible module, listed in lexicographic order.
The root system may be computed by the same procedure used to
determine the weight systems. To do this, of course, it is necessary to know
the highest root of the Lie algebra, which is listed in Fig. 12.
3.9 Freudenthal's Algorithm 109
where 6 - \ J>0 a denotes half the sum of the positive roots. The sum over
a goes over all positive roots, which are known from the root system. The
sum over j is not really infinite but breaks off after a finite number of terms.
3.10 The Weyl Character Formula 111
To use this formula to compute the multiplicities, we start with the multi-
plicity of the highest weight A, which we know is unity for an irreducible
module. If ^ is a weight other than the highest weight, and if we know the
multiplicities for all weights // > //, then each term on the right side of the
preceding expression is known. Since for any weight /j. ^ A, the quantity
is nonzero and positive, we can always solve for n^ . The use of the Freudenthal
formula can be a lengthy process, but it can be done automatically on a
computing machine [4], [26]. Sometimes Freudenthal's algorithm may be
made more efficient by making limited use of Weyl reflections and by using
the fact that weights related by Weyl reflections have the same multiplicity
[148].
As an example of the use of Freudenthal's formula, we consider the
simple Lie algebra B2 . We have already computed the metric :
The positive roots are a l 5 a 2 , a: + a 2 and 2a.1 + <x 2 , and half their sum is
5 = A! + A 2 . Now consider again the 16-dimensional module over B2 with
highest weight A = Aj + A2 which we previously discussed. Its weight
system consists of an octagon with a square inside it. Since the eight weights
of the octagon are related to the highest weight by the Weyl group, these
must have the same multiplicity as the highest weight, that is, one. The four
weights of the square inside the octagon are also related to each other by
Weyl reflections, and we thus need only compute the multiplicity of one of
these. Applying the Freudenthal formula to the case we calculate
The Weyl formula can be proved by purely algebraic means, and has recently
also been extended to algebraic groups [48], [218].
112 Constructive Methods 3.10
where 6 is half the sum of the positive roots. This quantity 5 is also equal
to the sum of the basic weights :
We define an antisymmetrizer
where 6 is half the sum of the positive roots, and the Weyl formula may be
written as
Since the elements e* form a basis for the group algebra, if the characters
of two irreducible modules are equal, then their highest weights are equal,
and hence the modules are isomorphic. Any finite-dimensional module M
over a semisimple Lie algebra is completely reducible. This means that M
114 Constructive Methods 3.11
Since the primitive characters are linearly independent, the m A are uniquely
determined, and to this extent at least the reduction of M is unique.
This procedure allows us to relate the /l-girdle to the 0-girdle since one can
show that
Thus we can obtain ;,(thd) for small t from the corresponding formula for
0(th8) simply by replacing S by A + <5, giving
3.11 The Weyl Dimension Formula 115
This yields the Weyl formula for the dimension of the irreducible module
with highest weight X :
and the formula for the girdle reduces to two terms. Also, wa/^ = /i, and
hence,
Thus \}/i is defined almost exactly like the character &, except that n^ is
replaced by unity for all weights fi belonging to the weight system A. Thus
if/1 can be calculated from a knowledge of the weight system alone, without
knowledge of the multiplicities of the weights.
The weight system in general has a hexagonal-type shape, as shown in
Fig. 13, and may be obtained from the highest weight A = pAj 4- qX2 by
the process of Weyl reflection and filling in ladders.
The girdles of the simple Lie algebra A2 satisfy the recursion relation
from which we may obtain by iteration a formula for the character [7] . For
triangular weight systems (those with either p or q equal to zero) we have
while for nontriangular weight systems, the recursion relation yields the
formula
FIG. 13. Weight diagram of the 105-dimensional representation D(6, 2) of A2, showing the
multiplicity of each weight
3.13 The Kostant and Racah Character Formulas 117
simple Lie algebras A3 % D3, B3 and C3 over the complex numbers. Some
early calculations of weight diagrams for the rank three simple Lie algebra
A3, corresponding to the Lie groups 5(7(4) and SO(6, R), were given in a
paper by Wigner over thirty years ago [247]. Some examples of weight
diagrams for rank three simple Lie algebras have been given by Loupias,
Sirugue and Trotin [154]. The most complete hand calculations for rank
three were carried out by Konuma, Shima and Wada [143]. For higher rank
algebras, fewer results have been reported in the literature. Examples of
weight multiplicity tables for the exceptional simple Lie algebra F4 are given
by Dynkin and Veldkamp [83], [231] . Speiser has given the weight diagrams
for the basic modules of Lie algebras of the types At, Bl and >r(see [216]).
Several other multiplicity tables have also appeared in the literature [90],
[218] . The difficulty of the hand computations led to considerable discussion
of improvements in computational methods [109], [110].
Kostant gave another character formula which is closely related to the
Weyl character formula. The Kostant formula provides a closed form
expression for the multiplicity n^ for any weight fj. of the irreducible module
with highest weight A over a semisimple Lie algebra [145], [146]. In a sense,
the Kostant formula is obtained from the Weyl character formula by
formally carrying out the girdle division. To do this, one uses a partition
function P(/i) defined for all linear forms // in H% as the number of ways of
writing /z as a linear combination of positive roots with nonnegative integers
as coefficients. That is, P(fi) is the number of ways of writing
Racah [196] suggested another formula for characters and this method
has also been described by Klimyk [140]. The Racah formula provides a
recursion relation directly for the multiplicities of the weights:
Given the weight diagrams, this formula could be used directly to compute
the outer multiplicities by a process of repeated subtraction. The leading
term on the right is the one for
Subtracting this term from both sides and examining the leading terms of
the remaining expression yields the outer multiplicity of the next highest
submodule, and so on. This procedure can easily be programmed for a
computing machine, and thus each method for computing weight diagrams
yields a corresponding method for computing Clebsch-Gordan series. In
practice, this procedure for computing the outer multiplicities can be made
more efficient [140], [155]. For example, if the Weyl girdle method is chosen,
only one girdle division is needed if we rewrite the above formula as
120 Constructive Methods 3.14
Some applications of the Steinberg formula have been given in the literature
[198], [221], [222].
The Steinberg formula is a generalization of the original Clebsch-
Gordan formula for computing the tensor product of modules over the
simple Lie algebra A 1 . The character of the irreducible module M^ with
spin j, highest weight A = jet, and dimension 2j + 1 is
We may use this character formula directly to reduce the tensor product
M A (g) Mr of two such modules as a direct sum of irreducible submodules.
Setting /I = ja and A' = /a, we find that the character of the tensor product
module is
For AI, the partition function P(n) is unity if /* = na, where n is an integer
^0, and zero otherwise. Clearly m A Al ^ 2 = 0 unless jl + j2 j is an integer.
Since 7, ^ 0, j2 ^ 0 and 0 ^ ; ^ ^ + j2 , the last term in the Steinberg
formula is always zero. Of the two terms with negative signs, only one can
contribute at a time, and it does so when \jl - j2\ > j, canceling out the
contribution from the first term on the right. Thus we find that m*^^ = 1
when y'i + j2 - j is an integer and l^ - j2\ ^j^ Ji + J2, while m\ lA2 = 0
otherwise.
3.14 The Steinberg and Racah Formulas 121
We shall also illustrate the use of the Steinberg formula for the simple
Lie algebra A2, corresponding to the group S U(3). The Cartan matrix of A2 is
Hence the basic weights Aj and A2 are related to the simple roots o^ and a 2 by
From the root system, shown in Fig. 4 ( 2.15), we see that cil , a 2 and a t + a2
are the positive roots.
Using the Weyl formula, we find that the dimension of an irreducible
module with highest weight plv + qk2 is given by
In particular,
and
or in other words,
To apply the Steinberg formula, of course, we need to use the Weyl group.
The Weyl group W of the simple Lie algebra A2 consists of the elements 1,
w
i > W 2> W i w 2> vv 2 W! and WIH^W, = w 2 WiW 2 . This group is isomorphic to
the symmetric group S3 of all permutations on three elements, the corre-
spondence being given by
The action of the Weyl group on the positive roots is shown in Fig. 14
122 Constructive Methods 3.15
These results could also have been obtained by other means, besides the
Steinberg formula. For this Lie algebra, and more generally also for the
simple Lie algebras of types A-D, there are some frequently used alternative
methods involving Young tableaux for computing Clebsch-Gordan series
[115].
Another formula for computing Clebsch-Gordan series is due to
Racah. It is a recursive formula which avoids the partition function and the
double summation over the Weyl group, but does require the computation of
the weight diagram of the module M^ with highest weight A t . Racah' s
formula is
The simple Lie algebra Al may be identified with the linear Lie algebra
consisting of all traceless linear operators in an (/ + l)-dimensional vector
space M over the complex number field. The vector space M may itself be
regarded as an irreducible module with highest weight ^, and may thus be
identified with the basic module M j . The module M is an elementary
module over the simple Lie algebra At, and the other basic modules M 2 ,
M 3 , , M, can be constructed as exterior powers of M. Explicitly, the
124 Constructive Methods 3.16
FIG. 16. Elementary modules for tensor analysis in the simple Lie algebras Bt,Ct, Dt
For the orthogonal algebras B, and >,, the exterior powers A * M again yield
all but one or two of the basic modules. The omitted modules are the basic
spinor module of B, and the two basic semispinor modules of D{, which we
shall discuss later. For the symplectic Lie algebras C,, the exterior powers
A * M are not irreducible, and to obtain the basic modules this method must
be supplemented by a trace removal process.
Once the basic modules have been constructed from the elementary
ones, we can then construct all irreducible modules via Cartan composition,
making use of the enveloping algebra. In this way, in principle, the entire
representation theory is reduced to a single elementary module. For some
applications it is advantageous to bypass the Cartan composition procedure
and construct the irreducible modules directly from the elementary module
via tensor analysis. To do this, one needs to study the symmetrization of the
tensor powers of the elementary module M. We consider here the case of
algebras of type Al for which this procedure is fairly straightforward [115],
[131]. The tensor power module (X)"M may be considered not only as a
module over the Lie algebra At, but also as a module over the permutation
group Sn. For each permutation n in Sn, there is a linear operator on the
3.16 Young Tableaux 125
tensor power module (X)"M which permutes the various factors in the
tensor product. We define the action of n on the tensor product of the vectors
xl5 , xninMby
The ideals of the group algebra of the permutation group can be labeled
by means of Young shapes as follows. A partition [a] of an integer n is a
list of integers [a x , , ak] such that al ^ a 2 ^ ^ a k > 0 and
A Young tableau is a Young shape in which the boxes have been num-
bered from 1, , n in some order. A standard Young tableau is a tableau
with the property that if all but the boxes labeled 1, , h with h < n are
erased, then these remaining h boxes form a Young tableau by themselves.
An example of a standard Young tableau is
The number fa of standard Young tableaux for a given Young shape [a]
Let a"s denote the permutation which transforms the tableau [a]s into
the tableau [<x]r :
3.16 Young Tableaux 127
and define
The elements Ears form a basis for the group algebra of the permutation group,
but do not have simple multiplication laws. To obtain simpler basis ele-
ments, we first expand E*s in terms of all the permutations n in Sn :
Let (a )~ l = rja denote the matrix inverse to the matrix whose elements are
the coefficients gs. Then the elements
form a basis of the group algebra of the permutation group, and these have
the simple multiplication properties mentioned earlier. It turns out that
the matrix rf is in fact very close to being a unit matrix, and for the elements
with r s, with which we are primarily concerned here, these coefficients
effectively disappear, and we have simply
Thus the projection operators can be computed directly from the positive
and negative groups of the tableau, without requiring a calculation of the
>tfs coefficients [204].
The permutations n in Sn commute with the elements x in the Lie
algebra and hence so do the elements e*> from which it follows immediately
that e*Mn is a submodule of the module M". The modules e*M" for the
same Young shape a, but different tableaux r and s, are isomorphic :
For algebras of type A{ , these modules are all irreducible, and each irreducible
module can be obtained in this way by projecting out symmetrized portions
of the tensor powers of the elementary module [244] . In general, however,
there will be infinitely many different Young shapes corresponding to a given
irreducible module. To obtain a one-to-one correspondence between
isomorphism classes of irreducible modules and Young shapes for Lie
algebras of type At, we may restrict the class of Young shapes under con-
sideration to those with at most / rows. The irreducible modules can then be
characterized either by such a restricted Young shape [a] or by the highest
128 Constructive Methods 3.17
weight L For a Young shape [a] = [o^ , , aj with a total of n boxes, the
highest weight of the modules e?M" is
where ^ , , A, are the basic weights and we set a,+ x = 0. Since the pro-
jection operators ef form a resolution of the identity, we also have
3.17 CONTRACTIONS
For Lie algebras of types Bh Ct and Dt> the reduction of the tensor
powers of a module M into symmetrized tensor powers e*Mn is not complete
because these submodules need not be irreducible [6]. To remedy this, we
need to discuss the process of contraction of tensors and the removal of
traces [195].
For the module M we shall take the space traditionally used to define
the algebras Bt, C{ and />,. It has 21 + 1 dimensions for the odd-type orthog-
onal Lie algebras Bt, and it has 21 dimensions for both the even-type orthog-
onal Lie algebras D, and the symplectic Lie algebras Ct. The defining
module M is equipped with a nonsingular bilinear form (x,y) which is
symmetric in the case of the orthogonal Lie algebras and antisymmetric
for the symplectic Lie algebras. In both cases, the Lie algebra elements a e L
are linear operators on M which are antisymmetric with respect to the
bilinear form
by
allows us to talk about orthogonality for tensors. One can now classify
tensors according to their degree of tracelessness, and thus arrive at the
orthogonal direct sum decomposition
Here M" consists of completely traceless tensors, that is, elements of M"
which are annihilated by any contraction. The space M"2 consists of tensors
which are orthogonal to all traceless tensors but which are annihilated by
a product of any two contractions and so forth, producing spaces Ml which
are mutually orthogonal and linearly independent. Finally, the remainder,
if any, is the subspace M tt0 orthogonal to all of the subspaces Ml with k
= 1, 2, , [n/2], which therefore consists of those tensors most resistant
against being annihilated by contractions.
As an example of such a decomposition into traceless tensors we may
consider an orthogonal Lie algebra. In the module M we may introduce an
orthonormal basis ev , , en satisfying
FIG. 17. Dynkin diagrams for some of the irreducible modules for the Lie algebra C3
The Dynkin diagrams for the complete weight system of the fundamental
6-dimensional module and the simple roots of C3 are given in Fig. 18.
The weight system here was calculated by using the Dynkin algorithm
discussed earlier (3.8).
The following Clebsch-Gordan series of modules over C3 can be
obtained, for instance, from a study of characters :
FIG. 18. Roots and fundamental weights for the simple Lie algebra C3
3.18 Spinor Analysis and Clifford Algebras 131
The Dynkin diagrams for the various modules which appear here are
among those included in Fig. 17.
The decomposition of the antisymmetrized tensor powers can now be
written down by inspection of the dimensions involved. The module
{6} {6} is a 36-dimensional vector space, which breaks up into a sym-
metric part with (6 x 7)/(l x 2) = 21 dimensions and an antisymmetric
part with (5 x 6)/(l x 2) = 15 dimensions. From these dimensional con-
siderations and the Clebsch-Gordan series decomposition given previously,
it is clear that the decomposition of the antisymmetric part is given by
and all higher powers are zero. From this we see that the basic modules of
C3 can still be constructed from the exterior powers of the fundamental
module M, but some additional process must be used to separate off the
accompanying one- and six-dimensional modules.
FIG. 19. Dynkin diagrams for the basic spinor and semispinor modules over orthogonal Lie
algebras B, and D,
is an associative algebra with unity called the first Clifford algebra of the
module M with respect to the bilinear form (x,y). Since the composition of
the canonical mapping of T(M) onto C(M) with the inclusion mapping of M
into T(M) yields a one-to-one mapping, we may identify M with its image
in C(M). The Clifford algebra C(M) obtained in this way satisfies the follow-
ing universality property. If A is an associative algebra with unity, then any
linear mapping y:M -> A which satisfies
for all x, y e M . The generators and relations for a Clifford algebra take the
especially simple form
for all z, z' e M. If we let Mk be the subspace of C(M ) spanned by all elements
of the form a(zt) a(zk) 1, then
and hence,
In other words, the second Clifford algebra is spanned by the even elements
of the first Clifford algebra, and it follows that it is a subalgebra with
For example, the first Clifford algebra for the Lie algebra Al w B1
corresponding to the ordinary rotation group S0(3, R) is generated by three
elements xltx2 and x3 satisfying the Jordan relations. In this case, the
first Clifford algebra C(M) is an eight-dimensional algebra spanned by the
elements I,x1,x2,x3,xlx2,x2x3,x1x3 and xlx2x3. The second Clifford
algebra C + (M) is the four-dimensional subalgebra of C(M) spanned by
I , / = x1x2, j = x2x3 and k = xlx3. Since ij and k satisfy i2 = j2
k2 = 1, and ij = ji = k, jk = kj= i, and ki = ik = j, we may
identify C + (M) as the complex quaternion algebra.
Since the Clifford algebra C(M) is an associative algebra, we may also
regard it as a Lie algebra C(M)L under commutation. To see that the Lie
algebra L itself may be regarded as a subalgebra of the Lie algebra C(M)L,
we argue as follows. For all x, y and z in M we may obtain the relation
from the Jordan relation by writing the term 4(y, z)x on the right-hand side,
for instance, as
then we have
3.18 Spinor Analysis and Clifford Algebras 135
for all x and y in M, and hence, f(a) e L. Since one may show that /: M 2 - L
is a one-to-one Lie algebra homomorphism, it follows that
Since the orthogonal Lie algebra L has the same dimension, dimL = ( 2 ),
the mapping / is onto, and this establishes the isomorphism of M2 with L.
The simply-connected Lie groups corresponding to the compact real
forms of BI and Dt are not the orthogonal groups S0(n, R), but their covering
groups Spin(n). To obtain these spin groups from the Clifford algebra,
we consider the group G of all invertible elements weC(M) such that
uzu~l eM for all ze M. The spin group Spin (ri) is then the identity com-
ponent of the subgroup of G consisting of all elements whose left regular
representation restricted to any ideal of C(M) has determinant one [54], [55].
To obtain the spinor and semispinor modules, we must distinguish the
case n = 21 + 1 with L = B, from the case n = 21 with L = D,. For odd n,
the first Clifford algebra C(M) is a semisimple associative algebra, isomorphic
to the direct sum of two copies of C + (M). The second Clifford algebra
C + (M) for odd n is a simple associative algebra, isomorphic to the algebra
lin (N) of linear operators on some vector space N. This vector space N
of dimension 2' will turn out to be the basic spinor module over Bl. We may
write C + (M) lin(N) N N*, and since the basic spinor module N
also happens to be self-dual, this simplifies to C + (M) N N. Thus, for
the case n odd, the second Clifford algebra C+(M) is isomorphic, as a
module over L, to the tensor square N N of the basic spinor module N.
For even n, the first Clifford algebra C(M) itself is simple, and thus is iso-
morphic to the algebra of linear operators on some vector space of dimension
2'. This vector space may be identified with the direct sum Nv + N2 of the
two basic semispinor modules N x and AT2 over >,, each having dimension
2'"1. While the semispinor modules themselves are self-dual only for even /
and each is the dual of the other for odd /, their direct sum Nv + N2 is
self-dual in either case. Thus, for even n we may write the first Clifford algebra
as
where N = N1 + N2. The second Clifford algebra C + (M) for even n is not
simple, but only semisimple, being the direct sum of two simple ideals,
namely,
particular basis for the fundamental module M over the orthogonal Lie
algebra L = Bt . There is a basis x0 , x j , , x / for M such that
and
we have
where / ! < < ir and ^ < <js. The second Clifford algebra then
factorizes as a product C+(M) = UV, where U and K are generated by the
w's and y's respectively, together with 1. The subspace N l/-( u i ty)
c= C + (M) is a 2'-dimensional left ideal of C + (M). Hence M2N <=. N, and
we may regard N as a module over M2 L.
We may illustrate this construction of the spinor module for the simplest
case, corresponding to the Lie algebra A^ B. The second Clifford algebra,
which in this case is the quaternion algebra, may be generated from the
elements
and
The algebra C + (M), spanned by !,, and MU, factorizes as 17F, where U
is spanned by 1 and u, while V is spanned by 1 and v. The two-dimensional
spinor module N = Uv is spanned by v and wy. The factorization of the
second Clifford algebra,
for the Lie algebra A t corresponds to the Clebsch-Gordan series {2} {2}
3.19 Tensor Operators 137
where p is the relative momentum, r is the distance between the sun and the
planet, m is their reduced mass, and k is a constant. Due to the rotational
invariance of the Hamiltonian, the angular momentum
A set of three dynamical variables V = (Vx, Vy, Vz) is called a vector dynamical
variable if the Poisson bracket relations
hold. The dot product of two vector dynamical variables is a scalar dynamical
variable, while their cross product is a vector dynamical variable. Since
r and p are vector dynamical variables, it follows immediately that L x p
is also a vector dynamical variable. Hence the Runge-Lenz vector is a vector
dynamical variable, and we have
For the bounded orbits, the system is invariant under a certain four-dimen-
sional rotation group, while the symmetry group appropriate for the
scattering trajectories turns out to be a Lorentz group. Since the energy is
negative for the bounded orbits, we have in this case simply
The vectors L and A together generate a Lie algebra which is the direct
sum of two ideals, each isomorphic to so(3, R). This algebra may be identified
with
Thus one of the simple ideals is generated by the elements J + and the other
by J~.
Essentially the same Kepler problem symmetry comes up again via
the correspondence principle in the quantum mechanical description of the
hydrogen spectrum. Here the problem is to solve the Schrodinger equation
140 Constructive Methods 3.19
because these operators do not commute. The SO(4, R) symmetry can be used
to analyze the degeneracy of the spectrum of atomic hydrogen, and
historically this was the way it was first done in quantum mechanics [187].
The idea is to use the two different Casimir operators, namely L2 + A2
and A L, of the Lie algebra S0(4, R). The Hamiltonian H is related to the
first Casimir invariant by
where we have chosen units in which h = 1. From the explicit formulas for
L and A, it is easy to see that the other Casimir invariant A L is always
zero. Each eigenspace of the hydrogen atom Hamiltonian is a module over
the Lie algebra 50(4, R), but not all modules are actually realized as such
eigenspaces.
In general, the irreducible modules over a given semisimple Lie algebra
can be constructed as tensor products of irreducible modules over its
simple ideals. Explicitly, if a Lie algebra L is the direct sum of two ideals
L = L! L2 and if M1 and M2 are irreducible modules over Ll and L2,
respectively, then we may regard M x M2 as an irreducible module over L.
3.20 Charge Algebras 141
For the hydrogen atom, the irreducible modules can be specified by the
eigenvalues of the Casimir operators (J + ) 2 and (J~) 2 . Since
only those modules occur for which these eigenvalues are the same. Since
and since
we find with n = 2j + 1,
theory has been stimulated by the needs of atomic, nuclear, and most
recently, also elementary particle physics. Since one may anticipate that in
the future there will also be many more areas in which these methods will
find application, it is likely that this field will continue to grow for some time
to come.
We discussed earlier how Lie algebras arise in the study of conservation
laws in classical and quantum mechanics. The Lie algebra of all conserved
quantities is usually infinite-dimensional since this algebra is closed not
only under the Lie product, but also under ordinary multiplication. One
can frequently obtain useful results by working with a finite-dimensional
subalgebra. For example, such finite-dimensional Lie algebras often arise
naturally as algebras of additive quantities. An additive quantity for a system
of particles is a quantity which is a sum of terms, one for each particle,
such that the term for a given particle depends only on the nature and
coordinates of that particle. Additive quantities in physics are generally
divided into two groups, those of kinematical origin, like energy and momen-
tum, and those which are not of kinematical origin. The latter are often
called generalized charges because the ordinary electric charge is an example
of just such an object. A conserved charge is one which does not depend
explicitly on the time and which commutes with the Hamiltonian. Besides
the electric charge, there is another generalized charge which is always
conserved, called the baryon number. Barring the existence of certain
hypothetical particles called quarks, the electric charge and baryon num-
ber of every known particle in physics is an integer, positive, zero or
negative.
It is widely believed that the generalized charges form a Lie algebra,
and that the set of conserved charges is a subalgebra of this Lie algebra.
Some evidence for this idea comes from the description of additive quantities
within the formalism of quantum field theory. Here we consider only the case
that all particles can be specified by means of a discrete label, which would
be the case, for instance, if they were located at fixed positions and not allowed
to move. One introduces for each type of particle a set of creation and annihi-
lation operators acting on the Hilbert space of states. The creation operators
a\, , al are required to be the Hermitian conjugates of the set of annihi-
lation operators al, , an. If \l/ is a vector in Hilbert space corresponding
to some quantum state, then the vector a]\l/ corresponds to another state
differing from the original state by the addition of a single particle of type /
Similarly, the vector afi corresponds to a state obtained by removing a
single particle of type j. These operators satisfy a certain canonical set of
commutation or anticommutation relations, depending on the spin of the
particle [43], [189].
For particles with integer spin, called bosons, they satisfy commutation
relations identical to those written down for harmonic oscillators:
3.20 Charge Algebras 143
One of the main reasons why the theory of harmonic oscillators is so funda-
mental to quantum mechanics is just because it underlies this formal
apparatus of second quantization. The creation and annihilation operators
are to a large extent determined by their commutation relations. Formally
one may regard the Hilbert space as a module over the oscillator algebra
defined by these commutation relations. If the whole Hilbert space can be
generated from a single vacuum state, for example, we obtain the Fok
representation of this oscillator algebra [88]. Von Neumann showed that
for n < oo, the Fok representation is the only irreducible representation
of the oscillator algebra [235]. For n < oo, every representation of the oscil-
lator algebra is a direct sum of Fok representations, but for n = oo the
situation becomes more complicated [51].
For half-integer spin particles, called fermions, these creation and
annihilation operators satisfy anticommutation relations,
where
thus verifying that the map C i-+ a^Ca is a Lie algebra homomorphism.
It would be desirable to have a more direct argument that additive
quantities form a Lie algebra without appealing to the full apparatus of
quantum field theory. We sketch such an argument, which is based on the
notion of asymptotic products of state vectors. In order to say definitely
that a system is made up of a certain number of particles, it is necessary
that these particles be separated far enough apart. Otherwise their inter-
actions may be strong enough to annihilate some of the particles or to create
new particles. The state vector of a well-separated composite system may
be described as a product of the state vectors of its component clusters [113].
144 Constructive Methods 3.20
It is then easily seen that the commutator of two additive quantities is again
an additive quantity.
Lie algebras of conserved generalized charges have recently been
studied in elementary particle physics. Elementary particles engage in four
distinct classes of interactions, known as the strong, electromagnetic, weak
and gravitational interactions [59] , [253] . The strong interactions possess
greater symmetry than the others, and it is here that Lie group theory enters
particle physics most directly. Not all particles engage in strong interactions ;
those that do are called hadrons, while those that do not include the leptons
(neutrino, electron and muon) and their antiparticles, as well as the photon
and the (hypothetical) graviton. The hadrons may be further classified by
their value of the baryon number. Hadrons having baryon number 0,
1 or - 1, for example, are called mesons, baryons and antibaryons, respectively.
All known mesons have integer spin, while all known baryons have half-
integer spin.
Because the underlying dynamics of hadrons is poorly understood,
the use of symmetry considerations has been particularly valuable. Experi-
mentally, strong interaction symmetries are manifested most strikingly in
their charge and mass spectra. The earliest evidence of this sort is the near
equality of the proton and neutron masses, leading to the theory of isotopic
spin [119]. The state vectors for the proton and neutron span a two-dimen-
sional module over SU(2), called the nucleon isotopic-spin doublet. The
isotopic spin operators themselves form a representation of the corre-
sponding Lie algebra. More recently, it was found that the SU(2) symmetry
can be extended to an approximate SZ/(3) symmetry [27], [100], [217]. The
proton and neutron, together with the lambda A, sigma E + ~ and xi
hyperons E S~,span an eight-dimensional module over Sl/(3), called the
baryon octet. The strong interactions do not preserve the SU(3) symmetry,
but break it in a way which allows many of the properties of hadrons to be
predicted. For example, the masses of the baryons are not equal, but satisfy
a simple mass formula
The derivation of relations like the hadron mass formulas for SU(3)
involves somewhat more complicated computational techniques, especially
methods for calculating Clebsch-Gordan coefficients. To illustrate these
methods of representation theory, we shall work out one example for the
group SC/(3). The interested reader should note that the computation of
SU(2) and SU(3) Clebsch-Gordan coefficients is widely discussed in the
recent literature, and both tables and computer programs are available
[35], [58], [60].
The eight-dimensional representation whose Dynkin diagram is given
as in Fig. 20 plays an important role in applications. The Clebsch-Gordan
series
Let us use primes to distinguish the three octet modules that are involved in
our computation of the reduction
By inspecting the weight systems, we can write down the general form for
the highest weight vector of the module (8"}:
The coefficients A, B and C can be determined by making use of the fact that
MI 2 is an extreme vector: e^u"^ = e2u'[2 = 0. We obtain, by a simple com-
putation,
and thus,
The remaining basis vectors of the module (8"} can be obtained by applying
the lowering algebra to this highest weight vector. For example, the equation
[/2' e i2l ~e\ allows us to write down
Using our formula for u"l2, we then can immediately calculate u'[, obtaining
149
150 Bibliography
[22] D. W. BARNES, Lattice isomorphisms of Lie algebras, J. Australian Math. Soc., 4 (1964),
pp. 470-475.
[23] , On Carton subalgebras of Lie algebras, Math. 2., 101 (1967), pp. 350-355;
Erratum, Math. Z., I l l (1969), p. 168.
[24] R. E. BECK AND B. KOLMAN, Generation of the Weyl group on a computer, J. Computa-
tional Phys., 7 (1971), pp. 346-353.
[25] , Computer approaches to the representations of Lie algebras, J. Assoc. Comput.
Mach., to appear.
[26] , A computer implementation of Freudenthaf s multiplicity formula, to be published.
[27] R. E. BEHRENDS, J. DREITLEIN, C. FRONSDAL AND B. W. LEE, Simple groups and strong
interaction symmetries, Rev. Modern Phys., 34 (1962), pp. 1-40.
[28] J. G. F. BELINFANTE, B. KOLMAN AND H. A. SMITH, An introduction to Lie groups and
Lie algebras, with applications, SIAM Rev., 8 (1966), pp. 11-46.
[29] , An introduction to Lie groups and Lie algebras, with applications. II: The basic
methods and results of representation theory, Ibid., 10 (1968), pp. 160-195.
[30] J. G. F. BELINFANTE AND B. KOLMAN, An introduction to Lie groups and Lie algebras,
with applications. Ill: Computational methods and applications of representation
theory, Ibid., 11 (1969), pp. 510-543.
[31] L. C. BIEDENHARN, JR., Wigner coefficients for the /?4 group and some applications,
J. Mathematical Phys., 2 (1961), pp. 433-441.
[32] L. C. BIEDENHARN, JR. AND W. J. HOLMAN, III, Complex angular momenta and the groups
SU(1, /) and SU(2), Ann. Physics, 39 (1966), pp. I^t2.
[33] G. BIRKHOFF, Lie groups isomorphic with no linear group, Bull. Amer. Math. Soc., 42
(1936), pp.882-888.
[34] , Hydrodynamics: A Study in Logic, Fact, and Similitude, Princeton Univ. Press,
Princeton, N.J., 1950.
[35] R. BIVINS, N. METROPOLIS, M. ROTENBERG AND J. K. WOOTEN, JR., The 3 j and 6 j
Symbols, The M.I.T. Press, Cambridge, Mass., 1959.
[36] A. BOREL, Topology of Lie groups and characteristic classes, Bull. Amer. Math. Soc.,
61 (1955), pp. 397-432.
[37] R. BRANTON, A generalization of the fundamental theorem ofprojective geometry, Doctoral
thesis, University of Pennsylvania, Philadelphia, 1970.
[38] F. BRICKELL AND R. S. CLARK, Differentiable Manifolds: An Introduction, Van Nostrand-
Reinhold, London, 1970.
[39] H. C. BRINKMAN, Applications of Spinor Invariants in Atomic Physics, North Holland,
Amsterdam, 1956.
[40] R. BROCKETT, System theory on group manifolds and coset spaces, SIAM J. Control,
10 (1972), pp. 265-284.
[41] G. BROWN, A remark on semisimple Lie algebras, Proc. Amer. Math. Soc., 15 (1964),
p. 518.
[42] F. BRUHAT, Lectures on Lie groups and representations of locally compact groups, Lecture
notes, Tata Institute of Fundamental Research, Bombay, India, 1958.
[43] N. BURGOYNE, On the connection of spin with statistics, Nuovo Cimento, 8 (1958), pp. 607-
609.
[44] J. E. CAMPBELL, On a law of combination of operators, Proc. London Math. Soc., 29
(1898), pp. 14-32.
[45] E. CARTAN, Projective groups which do not leave any flat manifold invariant, Bull. Soc.
Math. France, 41 (1913), pp. 53-96 (in French).
[46] , The finite and continuous real simple groups, Ann. Sci. Ecole Norm. Sup., 31
(1914), pp. 263-355 (in French).
[47] P. CARTIER, Finite groups generated by reflections, Seminaire Chevalley, vol. 2, expose
14, Ecole Normale Superieure, Paris, 1957/1958 (in French).
[48] , On H. WeyTs character formula, Bull. Amer. Math. Soc., 67 (1961), pp. 228-230.
Bibliography 151
[101] R. C. GLAESER AND B. KOLMAN, Lattice isomorphisms of solvable Lie algebras, J. Australian
Math. Soc., 10 (1969), pp. 266-268.
[102] R. GODEMENT, A theory of spherical functions. I, Trans. Amer. Math. Soc., 73 (1952),
pp. 496-555.
[103] R. H. GOOD, Properties of the Dirac matrices. Rev. Modern Phys., 27 (1955),
pp. 187-211.
[104] M. GOTO, On algebraic Lie algebras, J. Math. Soc. Japan, 1 (1948), pp. 29-^45.
[105] , Lattices of subalgebras of real Lie algebras, J. Algebra, 11 (1969), pp. 6-24.
[106] M. GOURDIN, Unitary Symmetries and Their Application to High-Energy Physics, North
Holland, Amsterdam, 1967.
[107] S. GREENSPAN AND R. D. RICHTMYER, Expansion of the Campbell-Baker-Hausdorff
formula by computer, Comm. Pure Appl. Math., 18 (1965), pp. 107-108.
[108] H. J. GROENEWOLD, On the principles of elementary quantum mechanics, Physica, 12
(1946), pp. 405-474.
[109] B. GRUBER, Lecture notes on multiplicities in the classical groups, Institute di Fisica
Teorica, Universita di Napoli, Naples, Italy, 1968, unpublished.
[110] , Recurrence relations for the multiplicities in the classical groups, J. Mathematical
Phys., 11 (1970), pp. 1783-1790.
[Ill] B. GRUBER AND H. J. WEBER, On the construction of weight diagrams for SO(5), Sp(4)
andG2, Proc. Roy. Irish Acad. Sect. A, 66 A (1968), pp. 31^40.
[112] K. W. GRUENBERG AND A. J. WEIR, Linear Geometry, Van Nostrand, Princeton, N.J.,
1967.
[113] R. HAAG, The framework of quantum field theory, Nuovo Cimento Suppl., 14 (1959),
pp. 131-152.
[114] P. R. HALMOS, Finite-Dimensional Vector Spaces, 2nd ed., Van Nostrand, Princeton,
N.J., 1958.
[115] M. HAMERMESH, Group Theory and its Applications to Physical Problems, Addison-
Wesley, Reading, Mass., 1962.
[116] HARISH-CHANDRA, On representations of Lie algebras, Ann. of Math. (2), 50 (1949),
pp. 900-915.
[117] , On the radical of a Lie algebra, Proc. Amer. Math. Soc., 1 (1950), pp. 14-17.
[118] F. HAUSDORFF, The symbolic exponential formula in group theory, Ber. Verh. Sachs.
Ges. Wiss. Leipzig. Math.-Phys. K.I., 58 (1906), pp. 19-^8 (in German).
[119] W. HEISENBERG, On the structure of the atomic nucleus. I, Z. Physik, 77 (1932), pp. 1-11
(in German).
[120] S. HELGASON, Differential Geometry and Symmetric Spaces, Academic Press, New York,
1962.
[121] K. HELMERS, Symplectic invariants and Flowers" classification of shell model states,
Nuclear Phys., 23 (1961), pp. 594-611.
[122] J. C. HERZ. Contribution to the algebraic theory of partial differential equations, Ann. Sci.
ficole Norm. Sup., 71 (1954), pp. 321-362.
[123] E. L. HILL AND J. M. JAUCH, On the problem of degeneracy in quantum mechanics, Phys.
Rev., 57 (1940), pp. 641-645.
[124] G. HOCHSCHILD, The Structure of Lie Groups, Holden-Day, San Francisco, 1965.
[125] W. C. HOFFMAN, The neuron as a Lie group germ and a Lie product, Quart. Appl. Math.,
25(1968), pp.423-440.
[126] H. HOPF, Vector fields in n-dimensional manifolds, Math. Ann., 96 (1926), pp. 225-250
(in German).
[127] H. HOPF AND W. RINOW, On the concept of a complete differential-geometric surface,
Comment. Math. Helv., 3 (1931), pp. 209-225 (in German).
[128] T. E. HULL AND L. INFIELD, The factorization method, Rev. Modern Phys., 23 (1951),
pp. 21-68.
[129] T. HUSAIN, Introduction to Topological Groups, W. B. Saunders, Philadelphia, 1966.
154 Bibliography
[130] E. L. INCH, Ordinary Differential Equations, Longmans, Green & Co., London, 1926;
reprinted by Dover Publications, New York, 1956.
[131] C. ITZYKSON AND M. NAUENBERG, Unitary groups: Representations and decompositions,
Rev. Modern Phys., 38 (1966), pp. 95-120.
[132] M. JACOB AND G. C. WICK, On the general theory of collisions for particles with spin,
Ann. Physics, 7 (1959), pp. 404-428.
[133] N. JACOBSON, Rational methods in the theory of Lie algebras, Ann. of Math. (2), 36
(1935), pp. 875-881.
[134] , Lectures in Abstract Algebra. Volume II: Linear Algebra, Van Nostrand, Prince-
ton, N.J., 1953.
[135] , Lie Algebras, Interscience Tracts in Pure and Applied Mathematics, No. 10,
Interscience, New York, 1962.
[136] , Lectures in Abstract Algebra. Volume HI: Theory of Fields and Galois Theory,
Van Nostrand, Princeton, N.J., 1964.
[137] A. JOSEPH, Derivations of Lie brackets and canonical quantization, Comm. Math. Phys.,
17 (1970), pp. 210-232.
[138] R. JOST, Poisson brackets (an unpedagogical lecture), Rev. Modern Phys., 36 (1964),
pp.572-579.
[139] R. E. KALMAN, Mathematical description of linear dynamical systems, SI AM J. Control,
1 (1963), pp. 152-192.
[140] A. U. KLIMYK, Multiplicities of weights of representations and multiplicities of representa-
tions ofsemisimple Lie algebras, Soviet Math. Dokl., 8 (1967), pp. 1531-1534.
[141] B. KOLMAN, Semi-modular Lie algebras, J. Sci. Hiroshima University Ser. A-I Math.,
29 (1965), pp. 149-163.
[142] , Relatively complemented Lie algebras, Ibid., 31 (1967), pp. 1-11.
[143] M. KONUMA, K. SHIMA AND M. WADA, Simple Lie algebras of rank 3 and symmetries of
elementary particles in the strong interactions, Progr. Theoret. Phys. (Japan) Suppl.,
28(1963), pp. 1-128.
[144] P. Koosis, An irreducible unitary representation of a compact group is finite-dimensional,
Proc. Amer. Math. Soc., 8 (1957), pp. 712-715.
[145] B. KOSTANT, A formula for the multiplicity of a weight, Proc. Nat. Acad. Sci. U.S.A.,
44 (1958), pp. 588-589.
[146] , A formula for the multiplicity of a weight, Trans. Amer. Math. Soc., 93 (1959),
pp. 53-73.
[147] , Lie group representations on polynomial rings, Bull. Amer. Math. Soc., 69(1963),
pp. 518-526.
[148] M. I. KRUSEMEYER, Determining multiplicities of dominant weights in irreducible Lie
algebra representations using a computer, BIT, 11 (1971), pp. 310-316.
[149] J. KUCERA, Solution in large of control problem x = (A(\ - v) + Bv)x, Czechoslovak.
Math. J., 16 (1966), no. 91, pp. 600-623.
[150] , Solution in large of control problem x - (Au + Bv)x, Ibid., 17 (1967), no. 92,
pp.91-96.
[151] E. E. LEVI, On the structure of finite and continuous groups, Atti Accad. Torino, 40 (1905),
pp. 3-17 (in Italian).
[152] L. H. LOOMIS, An Introduction to Abstract Harmonic Analysis, Van Nostrand, New York,
1953.
[153] J. D. LOUCK, Recent progress toward a theory of tensor operators in the unitary groups,
Amer. J. Phys., 38 (1970), pp. 3-42.
[154] G. LOUPIAS, M. SIRUGUE AND J. C. TROTIN, On simple Lie groups of rank 3, Nuovo Ci-
mento. 38 (1965), pp. 1303-1325.
[155] A. J. MACFARLANE, L. O'RAIFEARTAIGH AND P. S. RAO, Relationship of the internal and
external multiplicity structure of compact simple Lie groups, J. Mathematical Phys.,
8 (1967), pp. 536-546.
Bibliography 155
[183] Y. NE'EMAN, The symmetry approach to particle physics, Proc. International Conference
on Nuclear Structure, 1963, pp. 172-187; Reprinted in Gell-Mann and Ne'eman,
The Eightfold Way, W. A. Benjamin, New York, 1964, pp. 302-317.
[184} E. NORMAN AND J. WEI, Lie algebraic solution of linear differential equations, J. Mathe-
matical Phys., 4 (1963), pp. 575-581.
[185] , On global representations of the solutions of linear differential equations as a
product of exponentials, Proc. Amer. Math. Soc., 15 (1964), pp. 327-334.
[186] S. OKUBO, Note on unitary symmetry in strong interactions, Prog. Theoret. Phys., 27
(1962), pp. 949-966.
[187] W. PAULI, JR., On the hydrogen spectrumfrom the standpoint of the new quantum mechanics,
Z. Physik, 36 (1926), pp. 336-363.
[188] , Mathematical contributions to the theory ofDirac matrices, Ann. Inst. H. Poincare,
6 (1936), pp. 109-136 (in French).
[189] , The connection between spin and statistics, Phys. Rev., 58 (1940), pp. 716-722.
[190] , Continuous groups in quantum mechanics, Ergeb. d. Exakt. Naturw., vol. 37,
G. Hohler, ed., Springer-Verlag, Berlin, 1965, pp. 85-104 (Posthumous publication
of 1958 CERN Report).
[191] S. PERLIS, Theory of Matrices, Addison-Wesley, Cambridge, Mass., 1952.
[192] L. S. PONTRVAGIN, Topological Groups, 2nd ed., Gordon and Breach, New York, 1966.
[193] L. PUKANSZKY, On the Kronecker products of irreducible representations of the 2x2
real unimodular group. I, Trans. Amer. Math. Soc., 100 (1961), pp. 116-152.
[194] G. RACAH, Theory of complex spectra. IV, Phys. Rev., 76 (1949), pp. 1352-1365.
[195] , On the decomposition of tensors by contraction, Rev. Modern Phys., 21 (1949),
pp. 494-496.
[196] , Lectures on Lie groups, Group Theoretical Concepts and Methods in Elementary
Particle Physics, F. Gursey, ed., Gordon and Breach, New York, 1964, pp. 1-36.
[197] , Group theory and spectroscopy, Ergeb. d. Exakt. Naturw., vol. 37, G. Hohler,
ed., Springer-Verlag, Tracts in Modern Physics, Berlin, 1965, pp. 28-84 (Reprint
of 1951 lecture notes).
[198] D. RADHAKRISHNAN, On the Clebsch-Gordan series of a semisimple Lie algebra, J. Mathe-
matical Phys., 9(1968), pp. 2061-2063.
[199] D. RADHAKRISHNAN AND T. S. SANTHANAM, Internal multiplicity structure and Clebsch-
Gordan series for the exceptional group G(2), Ibid., 8 (1967), pp. 2206-2209.
[200] S. RAMANUJAM, Topology of classical groups, Osaka J. Math., 6 (1969), pp. 243-249.
[201] P. J. REDMOND, Generalization of the Runge-Lenz vector in the presence of an electric
field, Phys. Rev., 133 (1964), pp. 1352-1353.
[202] D. E. RICHMOND, Inverse square orbits: A simple treatment, Amer. Math. Monthly,
59 (1952), pp. 694-696.
[203] M. E. ROSE, Elementary Theory of Angular Momentum, John Wiley, New York, 1957.
[204] D. E. RUTHERFORD, Substitutional Analysis, Edinburgh University Press," Edinburgh,
1948.
[205] H. SAMELSON, Topology of Lie groups. Bull. Amer. Math. Soc., 58 (1952), pp. 2-37.
[206] T. SATAKE, On representations and compactifications of symmetric spaces, Ann. of Math.
(2), 71 (1960), pp. 77-110.
[207] R. D. SCHAFER, An Introduction to Nonassociative Algebras, Academic Press, New York,
1966.
[208] J. A. SCHOUTEN, Tensor Analysis for Physicists, Clarendon Press, Oxford, 1954.
[209] J. P. SERRE, Semisimple Complex Lie Algebras, W. A. Benjamin, New York, 1966 (in
French).
[210] J. R. SHEWELL, On the formation of quantum-mechanical operators, Amer. J. Phys., 27
(1959), pp. 16-21.
[211] D. J. SIMMS, Lie Groups and Quantum Mechanics, Lecture Notes in Mathematics, No. 52,
Springer-Verlag, Berlin, 1968.
Bibliography 157
[212] D. A. SMITH, Chevalley bases for Lie modules, Trans. Amer. Math. Soc., 115 (1965),
pp. 283-299.
[213] , A basis algorithm for finitely-generated Abelian groups, Math. Algorithms, 1
(1966), pp. 14-35.
[214] J. M. SOURIAU, Geometric Quantization, Comm. Math. Phys., 1 (1966), pp. 374-398
(in French).
[215] B. SPAIN, Tensor Calculus, University Math. Texts, No. 17, Oliver and Boyd, London, 1956.
[216] D. R. SPEISER, Fundamental representations of Lie groups, Helv. Phys. Acta, 38 (1965),
pp. 73-97.
[217] D. R. SPEISER AND J. TARSKI, Possible schemes for global symmetry, J. Mathematical
Phys., 4 (1963), pp. 588-612.
[218] T. A. SPRINGER, WeyVs character formula for algebraic groups, Inventiones Math., 5
(1968), pp. 85-105.
[219] R. STEINBERG, A general Clebsch-Gordan theorem, Bull. Amer. Math. Soc., 67 (1961),
pp. 401-407.
[220] E. L. STIEFEL, An Introduction to Numerical Mathematics, Academic Press, New York,
1963.
[221] N. STRAUMANN, On the Clebsch-Gordan series of semisimple Lie algebras, Helv. Phys.
Acta, 38 (1965), pp. 56-64.
[222] , Branching rules and Clebsch-Gordan series of semisimple Lie algebras, Ibid.,
38 (1965), pp. 481-498.
[223] R. F. STREATER, Canonical quantization, Comm. Math. Phys., 2 (1966), pp. 354-374.
[224] M. E. SWEEDLER, Hopf Algebras, W. A. Benjamin, New York, 1969.
[225] J. D. TALMAN, Special Functions: A Group Theoretic Approach, W. A. Benjamin, New
York, 1968.
[226] J. TARSKI, Partition function for certain simple Lie algebras, J. Mathematical Phys., 4
(1963), pp. 569-574.
[227] U. UHLHORN, Representation of symmetry transformations in quantum mechanics, Ark.
Fys., 23 (1963), pp. 307-340.
[228] B. L. VAN DER WAERDEN, The classification of simple Lie groups, Math. Z., 37 (1933),
pp. 446-462.
[229] V. S. VARADARAJAN, On the ring of invariant polynomials on a semisimple Lie algebra,
Amer. J. Math., 90(1968), pp. 308-317.
[230] O. VEBLEN AND J. H. C. WHITEHEAD, The Foundations of Differential Geometry, Cam-
bridge Tracts in Mathematics and Mathematical Physics, No. 29, Cambridge Uni-
versity Press, Cambridge, 1932.
[231] F. D. VELDKAMP, Representations of algebraic groups of type F+ in characteristic 2,
J. Algebra, 16 (1970), pp. 326-339.
[232] N. YA. VILENKIN, On the theory of associated spherical functions, Dokl. Akad. Nauk
SSSR, 111 (1956), pp. 742-744 (in Russian).
[233] , Bessel functions and representations of the group of Euclidean motions, Uspekhi
Mat. Nauk, 11 (1965), pp. 69-112 (in Russian).
[234] , Special Functions and the Theory of Group Representations, Amer. Math. Soc.
Mathematical Monograph Translations, No. 22, Amer. Math. Soc., Providence,
R.I., 1968.
[235] J. VON NEUMANN, The uniqueness of the Schrodinger operators, Ann. Math. Pura Appl.,
104 (1931), pp. 570-578 (in German).
[236] J. WEI, Note on the global validity of the Baker-Hausdorff and Magnus theorems, J. Mathe-
matical Phys., 4 (1963), pp. 1337-1341.
[237] L. WEISNER, Group-theoretic origin of certain generating functions, Pacific J. Math.,
5(1955), pp. 1033-1039.
[238] H. WEYL, The Concept of a Riemann Surface, first ed., B. G. Teubner, Leipzig, 1913
(in German).
158 Bibliography
A Basic weight, 93
Abelian group basis algorithm, 147 Bessel function, 72
Abelian ideal, 45 Bilinear form, 7
Abelian Lie algebra, 34, 44 antisymmetric, 8
Abelian Lie group, 34 matrix of, 8
Additive quantity, 142, 144 nonsingular, 9
Adjoint, 27 symmetric, 8
Adjoint representation, 52 Bilinear mapping, 40
Ado theorem, 53 Bilinear product, 12
Affine Lie algebra, 36, 44, 65 Boson, 142
Algebra, 12 Broken symmetry, 141
Algebraic group, 15
Analytic action, 20
Analytic function, 14 C
Analytic homomorphism, 32 Canonical coordinates, 26
Analytic manifold, 14 Cartan composition, 93
Analytic representation, 54 Cartan criterion, 53
Analytically related, 14 Cartan decomposition, 82
Angular momentum, 86, 97, 120 Cartan matrix, 73, 93
Anisotropic vector, 8 inverse of, 108
Annihilation operator, 30, 142 Cartan subalgebra, 66
Annihilator of module, 92 Casimir algebra, 98
Anticommutative product, 13 Casimir element, 72, 140
Anti-Hermitian matrix, 12 second order, 98
Antilinear operator, 27 Cayley numbers, 79
Antisymmetric bilinear form, 8 Character of module, 101, 102
Antisymmetrization, 60 computation, 110
Antisymmetrizer, 113, 125, 126 primitive, 113
Antiunitary operator, 27 Charge algebra, 143
Arc, 33 Charge multiplet, 144
Arcwise connected, 33 Chart, 14
Associative algebra, 12 Chevalley basis, 147
Atlas, 14 Circle group, 35
Automorphism of Lie algebra, 45 Classical mechanics, 24
Clebsch-Gordan coefficient, 137, 145-146
B Clebsch-Gordan series, 58, 102, 119, 120
Baker-Campbell-HausdorfF formula, 48, 52, Clifford algebra,
73 first, 132
Baryon number, 142 second, 133
Baryon octet, 144 Closed subgroup, 32
Basic module, 93 Coalgebra, 65
159
160 Index
G Inversion operation, 7
Gamma matrix, 132 Invertible linear operator, 15
Carding domain, 29 Involutive automorphism, 82
Gelfand-Zetlin basis, 147 Irreducible module, 56
General linear group, 5, 11, 22 Isomorphic
Generalized charge, 142 modules, 55
Generalized coordinates, 25 vector spaces, 43
Generalized momentum, 25 Isotopic spin, 144
Generalized simultaneous eigenvector, 67 Isotropic vector, 8
Girdle, 110, 112 Isotropy subgroup, 47
Girdle division, 112, 119
Global structure of Lie group, 32-34 J
Goldie-Ore theorem, 64 Jacobi identity, 13
Graded algebra, 59 Jordan relations, 132
Grassmann algebra, 60
Group algebra, 112, 125
K
Kepler problem, 138
H Kernel, 32
Hadron, 144 Killing form 53, 104
Half sum of positive roots, 112 Kostant formula, 118
Hamiltonian, 25 Kronecker delta, 9
Hamilton's equations, 25
Harmonic function, 70 L
Harmonic oscillator, 30, 71 Ladder generating Lie subalgebra, 70
Helicity representation, 88 Ladder of roots, 73
Hermitian operator, 27 Ladder of weights, 70
Highest root, 109 Lagrange's equations, 25
Highest weight, 74 Lagrangian, 25
Hilbert space, 26 Laplacian operator, 71
Homogeneous Lorentz group, 88 Lattice, 42
Homomorphism of ideals, 44
of Lie algebras, 32 Law of energy conservation, 26
of Lie groups, 32 Layer, 107
of modules, 55, 137 Left-invariant vector field, 19
of vector spaces, 43 Legendre transformation, 25
Hydrogen atom, 140 Leibniz rule, 52, 58
Hyperbolic basis, 10 Length of root, 76
Level, 69
I Levi decomposition, 45
Ideal, 43, 60 Lexicographic ordering, 69, 108
Idempotent operator, 125 Lie algebra, 12-13
Identity component, 33 ideal of, 43
Improper orthogonal matrix, 6 linear, 53
Indecomposable module, 56 module over, 54
Indefinite metric, 10 nilpotent, 65
Infinitesimal group, 18 of Lie group, 15, 18,23
Inhomogeneous Lorentz group, 88 radical of, 45
Integer matrix inversion, 104 semisimple, 45
Integer-mode arithmetic, 103 simple, 45
Integrating factor, 20 solvable, 44
Internal direct sum, 43 structure constants of, 18
Invariant solutions, 20 structure of, 46
Invariant subspace, 56 subalgebra of, 43
162 Index